[{"_1":2,"_994":-5,"_995":-5},"loaderData",{"_3":4,"_22":23,"_24":25,"_724":725},"root",{"_5":6,"_7":8},"toasts",[],"visitor",{"_9":10,"_11":12},"viewportWidth",1280,"visitorSession",{"_13":14,"_16":17,"_18":-5,"_19":20,"_21":-5},"attributes",[15],"BOT","id",7058884,"membership","referenceId","01971406-43e3-74e8-b880-2f000258007a","userAccount","routes/_public/_layout",{},"routes/_public/mcp/servers/~namespace/~slug/_pages/_layout",{"_26":27,"_82":-5,"_83":40,"_29":84,"_643":644,"_723":273},"availableLocales",[28,40,47,54,61,68,75],{"_29":30,"_33":37,"_38":39},"mcpServer",{"_31":32,"_33":34,"_35":36},"description","Bietet KI-Assistenten verbesserte Denkfähigkeiten durch strukturiertes Denken, persistenten Wissensgraphenspeicher und intelligente Tool-Orchestrierung zur Lösung komplexer Probleme.","name","MCP-Denkfabrik","readme","# MCP-Denkfabrik\n\n[![npm-Version](https://img.shields.io/npm/v/mcp-think-tank.svg?style=flat-square)](https://www.npmjs.com/package/mcp-think-tank) [![Lizenz: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![Typoskript](https://img.shields.io/badge/TypeScript-5.8-blue.svg)](https://www.typescriptlang.org/) [![Cursor-kompatibel](https://img.shields.io/badge/Cursor-Compatible-4B8DF8?logo=cursor\\&logoColor=white\\&style=flat-square)](https://www.cursor.so/) [![Claude-Kompatibilität](https://img.shields.io/badge/Claude-Compatible-9370DB.svg)](https://www.anthropic.com/) [![Modellkontextprotokoll](https://img.shields.io/badge/MCP-Tank-orange.svg)](https://github.com/modelcontextprotocol) [![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/) [![Schmiedeabzeichen](https://smithery.ai/badge/@flight505/mcp-think-tank)](https://smithery.ai/server/@flight505/mcp-think-tank)\n\n## Überblick\n\nMCP Think Tank ist ein leistungsstarker Model Context Protocol (MCP)-Server, der die Fähigkeiten von KI-Assistenten wie **Cursor** und **Claude @Web** erweitert. Er bietet eine strukturierte Umgebung für verbessertes Denken, persistenten Speicher und verantwortungsvollen Tool-Einsatz.\n\nZu den wichtigsten Funktionen gehören fortschrittliches **sequentielles Denken und verkettetes Denken** , ein robustes **Knowledge Graph Memory-** System mit Versionierung und intelligente **Tool-Orchestrierung mit Call-Limit-Schutzmechanismen** . Diese Plattform ermöglicht es KI, komplexe Probleme durch strukturierte Analysen zu lösen, Wissen über Sitzungen hinweg aufrechtzuerhalten und externe Ressourcen wie die Websuche zu nutzen – und das alles unter Einhaltung konfigurierbarer Nutzungsbeschränkungen.\n\n## 🎯 Philosophie\n\nMCP Think Tank basiert auf drei Kernprinzipien:\n\n1. **Elegante Einfachheit** : Minimale, gut gestaltete Tools, die die KI-Funktionen ergänzen, anstatt zu versuchen, sie zu replizieren.\n2. **Verbesserte Reflexion** : Sanfte Anleitung fördert besseres Denken und Selbstreflexion ohne starre Einschränkungen.\n3. **Persistenter Kontext** : Ein einfaches, aber effektives Wissensdiagramm bietet Gedächtnis für Gespräche.\n\n## Hauptmerkmale\n\n* 💭 **Denkwerkzeug** : Spezieller Bereich für strukturiertes Denken und Selbstreflexion.\n* 🧩 **Wissensgraph** : Einfaches und effektives persistentes Speichersystem.\n* 📝 **Tools zur Aufgabenverwaltung** : Planen, verfolgen und aktualisieren Sie Aufgaben, integriert in das Wissensdiagramm.\n* 🌐 **Web-Recherche-Tools (Exa)** : Durchsuchen Sie das Web und erhalten Sie mithilfe der Exa-API Antworten aus Quellen.\n* 🔍 **Memory Tools** : Benutzerfreundliche Tools zum Speichern und Abrufen von Informationen aus dem Wissensgraphen.\n* 🤝 **Client-Support** : Nahtlose Integration mit Cursor, Claude @Web und anderen MCP-Clients.\n* 🛡️ **Tool-Orchestrierung und Anruflimits** : Integrierte Sicherheitsvorkehrungen für eine effiziente und verantwortungsvolle Tool-Nutzung mit konfigurierbaren Limits.\n* ⚡ **Inhalts-Caching** : Leistungsoptimierung für Datei- und URL-Operationen mit automatischer Duplikatserkennung.\n* 🔄 **Sequentielles Denken** : Ermöglicht mehrstufige Denkprozesse mit Fortschrittsverfolgung.\n* 🔎 **Selbstreflexion** : Automatisierte Reflexionsdurchläufe zur Verbesserung der Argumentationsqualität.\n* 📊 **Strukturierte Ausgaben** : Automatische Formatierung von Denkprozessen für bessere Lesbarkeit.\n* 🔗 **Forschungsintegration** : Nahtlose Einbindung von Ergebnissen aus der Webrecherche in Argumentationsabläufe.\n\n## Vorteile des strukturierten Denkens\n\nDurch die Nutzung des `think` wird ein spezieller Raum für systematisches Denken geschaffen, der Folgendes fördert:\n\n* Klare Problemdefinition\n* Relevante Kontexterfassung\n* Schritt-für-Schritt-Analyse\n* Selbstreflexion über das Denken\n* Wohlgeformte Schlussfolgerungen\n\nAktuelle Studien zeigen deutliche Verbesserungen durch strukturiertes Denken:\n\n* **54 % relative Verbesserung** bei komplexen Entscheidungsaufgaben.\n* **Verbesserte Konsistenz** über mehrere Versuche hinweg.\n* **Verbesserte Leistung** bei Software-Engineering-Benchmarks.\n\n## Detaillierte Funktionen\n\nÜber die Kernliste hinaus bietet MCP Think Tank ausgefeilte Funktionen für erweiterte KI-Interaktion.\n\n### Strukturiertes Denken (Denkwerkzeug)\n\nDas `think` ist der zentrale Mechanismus für fortgeschrittenes KI-Denken. Es bietet eine dedizierte, strukturierte Umgebung, in der die KI Probleme systematisch zerlegen, Kontext erfassen, Optionen analysieren und Selbstreflexion durchführen kann. Dies ermöglicht eine tiefere Analyse und qualitativ hochwertigere Ergebnisse im Vergleich zu unstrukturierten Antworten. Es unterstützt sequentielle Schritte und lässt sich nahtlos in Recherche- und Gedächtnistools integrieren.\n\n#### Selbstreflexionsfunktion\n\nDas Think-Tool enthält eine leistungsstarke Selbstreflexionsfunktion, die mit dem Parameter `selfReflect: true` aktiviert werden kann:\n\n```javascript\nmcp_think-tool_think({\n structuredReasoning: \"...\",\n selfReflect: true,\n reflectPrompt: \"Optional custom reflection prompt\"\n})\n```\n\nBei aktivierter Selbstreflexion erhält die KI die Aufforderung, über ihre eigenen Überlegungen nachzudenken. Dies folgt der MCP-Designphilosophie, KI-Fähigkeiten zu verbessern, anstatt sie zu ersetzen.\n\nMit dem Parameter `reflectPrompt` können Sie die Eingabeaufforderung für die Reflexion anpassen und sie an bestimmte Denkaufgaben oder Domänen anpassen. Wenn dieser Parameter nicht angegeben ist, wird eine Standardeingabeaufforderung verwendet, die nach Inkonsistenzen, logischen Fehlern und Verbesserungsvorschlägen fragt.\n\n### Wissensgraph-Speicher\n\nDer Wissensgraph bietet einen persistenten Speicher über verschiedene Interaktionen und Sitzungen hinweg. Er ermöglicht der KI, ein wachsendes Verständnis des Projekts, seiner Komponenten und der damit verbundenen Konzepte aufzubauen.\n\n* **Mit Zeitstempel versehene Beobachtungen** : Alle Speichereinträge enthalten Metadaten zur Nachverfolgung.\n* **Duplikatsvermeidung** : Intelligentes Entity-Matching vermeidet redundante Einträge.\n* **Automatische Verknüpfung** : Heuristikbasierte Beziehungserstellung verbindet verwandte Konzepte (konfigurierbar).\n* **Erweiterte Abfragen** : Filtern Sie Ihren Speicher nach Zeit, Tags, Schlüsselwörtern und mehr mit dem leistungsstarken Tool `memory_query` für historische Analysen und die Nachverfolgung der Konzeptentwicklung. Finden Sie mühelos aktuelle Einträge der letzten 48 Stunden oder eines beliebigen Zeitraums.\n* **Speicherwartung** : Tools zum Beschneiden und Verwalten des Speicherwachstums sind enthalten.\n* **Wichtige Speichertools** : Tools wie `upsert_entities` , `add_observations` , `create_relations` , `search_nodes` , `memory_query` und `open_nodes` werden zur Interaktion mit dem Diagramm verwendet.\n\n### Tools zur Aufgabenverwaltung\n\nMithilfe einer Reihe von Tools kann die KI Projektaufgaben direkt im Gesprächsverlauf verwalten. Dadurch werden Planung und Ausführung in den Wissensgraphen integriert, sodass die KI den Projektstatus und die Prioritäten versteht.\n\n#### Wichtige Aufgabentools\n\n* `plan_tasks` : Erstellen Sie mehrere Aufgaben gleichzeitig mit Prioritäten und Abhängigkeiten\n* `list_tasks` : Aufgaben nach Status und Priorität filtern\n* `next_task` : Ruft die Aufgabe mit der höchsten Priorität ab und markiert sie als in Bearbeitung\n* `complete_task` : Aufgaben als erledigt markieren\n* `update_tasks` : Aktualisieren Sie vorhandene Aufgaben mit neuen Informationen\n\n### Web-Recherche-Tools (Exa)\n\nMCP Think Tank nutzt die Exa-API und bietet Tools zum Abrufen externer Informationen. Dadurch kann die KI auf aktuelle Informationen aus dem Internet zugreifen, um ihre Schlussfolgerungen zu untermauern und fundierte Antworten zu liefern.\n\n* `exa_search` : Führen Sie Websuchen basierend auf einer Abfrage durch.\n* `exa_answer` : Erhalten Sie eine präzise, mit Quellenangaben versehene Antwort auf eine sachliche Frage.\n\n**Hinweis:** Für die Verwendung dieser Tools müssen Sie Ihren Exa-API-Schlüssel konfigurieren. Weitere Informationen finden Sie im Abschnitt [„Konfiguration“](#configuration) .\n\n### Tool-Orchestrierung und Sicherheitsvorkehrungen\n\nMCP Think Tank umfasst umfassende Funktionen, um sicherzustellen, dass die Tools verantwortungsvoll und effizient eingesetzt werden.\n\n* **Nutzungslimits** : Eine konfigurierbare maximale Anzahl von Tool-Aufrufen pro Benutzerinteraktion ( `TOOL_LIMIT` , Standard: 25). Das Limit zählt nur aufeinanderfolgende Tool-Aufrufe innerhalb einer einzelnen Benutzernachricht und wird automatisch zurückgesetzt, wenn der Benutzer eine neue Nachricht sendet.\n* **Automatisches Tracking** : Alle Tool-Aufrufe werden protokolliert und überwacht.\n* **Graceful Degradation** : Wenn Grenzwerte erreicht werden, versucht das System, Teilergebnisse zurückzugeben.\n* **Intelligentes Caching** : Identische Tool-Aufrufe und wiederholte Datei-/URL-Inhaltsabrufe werden automatisch zwischengespeichert, was die Ausführungszeit und den Ressourcenverbrauch reduziert. Caching-Verhalten und -Größe sind konfigurierbar ( `CACHE_TOOL_CALLS` , `CONTENT_CACHE` ).\n* **Konfigurierbarer Zugriff** : Durch die Whitelist von Tools können verfügbare Tools in bestimmten Kontexten eingeschränkt werden.\n* **Fehlerbehandlung** : Eine robuste Fehlerbehandlung bietet klares Feedback bei Problemen wie dem Erreichen von Grenzwerten oder ungültigen Toolaufrufen.\n\n## 📦 Installation\n\n> ⚠️ **Wichtiger Hinweis LESEN SIE DIES:** Wenn Sie in Cursor oder Claude auf eine neue Version von MCP Think Tank aktualisieren, erstellen Sie möglicherweise mehrere Instanzen des MCP Think Tank-Servers, wodurch zusätzliche Node.js-Instanzen erstellt werden und die Systemleistung beeinträchtigt wird. Dies ist ein bekanntes Problem bei MCP-Servern. Beenden Sie alle MCP-Think-Tank-Prozesse in Ihrem System und überprüfen Sie, ob nur eine Node.js-Instanz ausgeführt wird.\n\n> ⚠️ Die Datei tasks.jsonl befindet sich in ~/.mcp-think-tank/. Die Datei ist von der kg-Datei getrennt, da der Think Tank durch zuvor erstellte Aufgaben in der kg-Datei verwirrt werden könnte. Löschen Sie den Inhalt der Datei tasks.jsonl, wenn die Datei zu groß wird oder wenn Sie ein neues Projekt starten und sicherstellen möchten, dass die Datei keine Aufgaben enthält. In einer zukünftigen Version werden Aufgaben möglicherweise mit der kg-Datei zusammengeführt, um sicherzustellen, dass abgeschlossene Aufgaben und Beziehungen im Speicher gespeichert bleiben und keine doppelten Aufgaben vorhanden sind.\n\n### NPX (empfohlen)\n\nDer einfachste Weg, MCP Think Tank zu verwenden, ist über NPX in Cursor mithilfe der Datei mcp.json, die die neueste Version ohne globale Installation ausführt.\n\n```bash\nnpx mcp-think-tank@2.0.7\n```\n\nFür die neueste Version (die möglicherweise Kompatibilitätsprobleme aufweist):\n\n```bash\nnpx mcp-think-tank@latest\n```\n\n> einige Benutzer haben Probleme mit npx @latest in Cursor. Versuchen Sie in diesem Fall, die Version mcp-think-tank@2.0.7 in der JSON-Datei anzugeben oder sie global zu installieren:\n\n### Globale Installation\n\nFür ein persistentes Befehlszeilentool:\n\n```bash\nnpm install -g mcp-think-tank\nmcp-think-tank\n```\n\n## ⚙️ Konfiguration\n\nMCP Think Tank wird hauptsächlich über Umgebungsvariablen oder über die Konfiguration Ihres MCP-Clients (wie Cursors `.cursor/mcp.json` ) konfiguriert.\n\n### Schnellstart: Grundlegende Einrichtung\n\n1. **Installieren Sie MCP Think Tank** (siehe Installation oben).\n2. **Holen Sie sich Ihren Exa-API-Schlüssel** (erforderlich für Websuchtools):\n * Melden Sie sich bei [exa.ai](https://exa.ai/) an und kopieren Sie Ihren API-Schlüssel.\n3. **WICHTIGE STDIO-SERVER SIND VERALTET** – Die MCP-Branche bewegt sich in Richtung HTTP-basierter Transporte. – ZUKÜNFTIGE UPDATES WERDEN STDIO-SERVER NICHT UNTERSTÜTZEN.\n4. **Konfigurieren Sie Ihren MCP-Server** (für Cursor fügen Sie ihn zu `.cursor/mcp.json` hinzu):\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"type\": \"streamable-http\",\n \"env\": {\n \"MEMORY_PATH\": \"/absolute/path/to/your/project/memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n }\n }\n }\n}\n```\n\n### Wesentliche Variablen\n\n* `MEMORY_PATH` : **Erforderlich** . Absoluter Pfad zur Speicherdatei. **Wichtig:** Legen Sie für jedes Projekt immer einen eindeutigen `MEMORY_PATH` fest, um Konflikte im Wissensgraphen zwischen Projekten zu vermeiden. Falls nicht angegeben, wird standardmäßig `~/.mcp-think-tank/memory.jsonl` verwendet.\n* `EXA_API_KEY` : **Erforderlich für Exa-Websuchtools** . Ihr API-Schlüssel von [exa.ai.](https://exa.ai/)\n\n### Erweiterte Konfiguration\n\n* `TOOL_LIMIT` : Maximale Anzahl an Tool-Aufrufen pro Benutzerinteraktion (Standard: `25` ). Der Zähler wird mit jeder neuen Benutzernachricht automatisch zurückgesetzt, sodass Sie innerhalb einer Interaktion bis zu 25 aufeinanderfolgende Tool-Aufrufe durchführen können.\n* `CACHE_TOOL_CALLS` : Aktivieren/Deaktivieren des Caching identischer Tool-Aufrufe (Standard: `true` ).\n* `TOOL_CACHE_SIZE` : Maximale Anzahl zwischengespeicherter Tool-Aufrufe (Standard: `100` ).\n* `CACHE_CONTENT` : Inhaltsbasiertes Caching für Datei-/URL-Operationen aktivieren/deaktivieren (Standard: `true` ).\n* `CONTENT_CACHE_SIZE` : Maximale Anzahl von Elementen im Inhaltscache (Standard: `50` ).\n* `CONTENT_CACHE_TTL` : Lebensdauer des zwischengespeicherten Inhalts in Millisekunden (Standard: `300000` – 5 Minuten).\n* `MCP_DEBUG` : Debug-Protokollierung aktivieren (Standard: `false` ).\n* `MCP_LISTEN_PORT` : Benutzerdefinierten Port für MCP-Server festlegen (Standard: `3399` für TCP-Server, nicht relevant für `stdio` ).\n* `LOG_LEVEL` : Protokollierungsebene festlegen ( `debug` , `info` , `warn` , `error` ) (Standard: `info` ).\n* `AUTO_LINK` : Aktiviert die automatische Entitätsverknüpfung im Wissensgraphen (Standard: `true` ).\n\n### Speicherwartung\n\n* `MIN_SIMILARITY_SCORE` : Schwellenwert für Entitätsübereinstimmung beim Verhindern von Duplikaten (Standard: `0.85` ).\n* `MAX_OPERATION_TIME` : Maximale Zeit für Batch-Speicheroperationen in Millisekunden (Standard: `5000` ).\n\nBeispielkonfiguration mit erweiterten Einstellungen in `.cursor/mcp.json` :\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"env\": {\n \"MEMORY_PATH\": \"./project-memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n \"TOOL_LIMIT\": \"50\",\n \"CACHE_CONTENT\": \"true\",\n \"CONTENT_CACHE_SIZE\": \"100\",\n \"MCP_DEBUG\": \"false\",\n \"AUTO_LINK\": \"true\"\n }\n }\n }\n}\n```\n\n> 💡 **Performance-Tipp:** Bei großen Projekten kann eine Erhöhung `TOOL_LIMIT` und Cache-Größen die Leistung verbessern, allerdings auf Kosten eines höheren Speicherverbrauchs. Beobachten Sie Ihre Nutzungsmuster und passen Sie diese entsprechend an. Im Cursor sollte das Tool-Limit jedoch 25 betragen, um ein Erreichen des Limits und die Fortsetzung vom letzten Tool-Aufruf zu vermeiden. Derzeit berichten viele Cursor-Nutzer von Problemen mit der Fortsetzung in Version 0.49.6. Dies hat nichts mit MCP Think Tank zu tun.\n\n> 💡 **Hinweis:** Wenn Sie Cursor im YOLO-Modus oder mit Vibe-Codierung verwenden, empfehle ich, neue Chats kontextbasiert vorzubereiten und Cursor mitzuteilen, dass er den MCP Think Tank zum Erstellen von Entitäten, Beobachtungen und Beziehungen verwenden soll. So holen Sie das Beste aus dem MCP Think Tank heraus.\n\nEin Beispiel für Kontext-Priming ist das Speichern einer `Prime.md` Datei im Ordner `.cursor` Ihres Projekts mit dem folgenden Inhalt:\n\n```markdown\n# Context Prime\n> Follow the instructions to understand the context of the project.\n\n## Run the following command\n\neza . --tree --git-ignore\n\n## Read the following files\n> Read the files below to get the context of the project. \n\n> list of files:\nREADME.md\n...\n\n## MCP Think Tank Tools\n> Test the MCP tools, first use 'show_memory_path' to remind the user of the current memory path file used, then use the 'memory_query' tool to find and read recent entities and observations for the last 48 hours so you are up to date.\n\n> Automatically utilize the MCP Think Tank to autonomously track project context, dynamically adding entities, observations, and relations to the knowledge graph while proactively querying for relevant information and historical insights. Use integrated planning and task management tools to enhance project efficiency. Keep track of the project and its context without the user having to ask for it.\n\n> Dont do anything else.\n```\n\nWeitere Einzelheiten zu MCP-Servern finden Sie in [der Cursor MCP-Dokumentation](https://docs.cursor.com/context/model-context-protocol) .\n\n## Projektregel-Setup (für Cursor/KI)\n\nUm sicherzustellen, dass Cursor und andere kompatible Agenten die Tools von MCP Think Tank effektiv nutzen, müssen Sie die KI anleiten. Dies geschieht üblicherweise über eine Projektregel. Erstellen Sie eine **einzelne Auto-Attach-Projektregel** wie folgt:\n\n### 1. Fügen Sie im Cursor eine neue Regel hinzu\n\n1. Cursor öffnen.\n2. Gehen Sie zur Befehlspalette ( `Cmd+Shift+P` oder `Ctrl+Shift+P` ).\n3. Wählen Sie **„Neue Cursorregel“** aus.\n4. Benennen Sie die Regel (z. B. `mcp-think-tank.mdc` ).\n5. Legen Sie im Regeleditor die Metadaten fest und fügen Sie den Regelinhalt aus dem folgenden Beispiel ein.\n\n### 2. Beispiel-Regeldatei ( `.cursor/rules/mcp-think-tank.mdc` )\n\nDiese Markdown-Datei dient *der KI* als Kontext und gibt ihr Hinweise, wann und wie die verfügbaren Tools zu verwenden sind.\n\n```mdc\nrule type: auto attach\nuse globs: **/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.md, **/*.py, **/*.json\n```\n\n\\----- Beginn der Regel -----\n\n```markdown\nRegularly utilize MCP Think Tank tools to maintain an updated knowledge graph and maximize its potential. Simply call the tools in your prompt. \n\n## Quick Decision Tree\n\n1. 🤔 **Complex problem to analyze?**\n → Use `think` to structure reasoning and reflect\n\n2. 🔍 **Need past context or information?**\n → Use `memory_query` (time-based) or `search_nodes` (keyword-based)\n\n3. 📊 **Planning implementation steps?**\n → Use `plan_tasks` to create and track work\n\n4. 🌐 **Need current external information?**\n → Use `exa_search` (general search) or `exa_answer` (factual questions)\n\n## Critical Memory Management (Automatic Use Required)\n\n| When to automatically use memory | Tool to use |\n|------------------|------------|\n| At session start | `memory_query` with recent time filter (last 24h) |\n| After completing significant analysis | `upsert_entities` to store conclusions |\n| When context seems missing | `memory_query` with relevant keyword |\n| Every ~30 minutes in long sessions | `upsert_entities` to create checkpoint |\n| When switching between major topics | `think` + `upsert_entities` to summarize progress |\n| Before session end | `upsert_entities` to store session summary |\n\n## Core Workflows\n\n### Workflow 1: Problem Analysis → Solution\n1. `memory_query` → Check for relevant past work\n2. `think` → Structure reasoning about the problem \n3. `plan_tasks` → Break down implementation steps\n4. `upsert_entities` → Store conclusions in memory\n\n### Workflow 2: Research → Knowledge Capture\n1. `memory_query` → Check if already researched\n2. `exa_search` → Find current information\n3. `think` → Analyze findings\n4. `upsert_entities` → Document key concepts\n\n### Workflow 3: Context Recovery (Session Resume)\n1. `memory_query` → Retrieve recent work (past 24-48h)\n2. `open_nodes` → Get details on relevant entities\n3. `think` → Synthesize context and plan next steps\n4. Continue where left off\n\n### Workflow 4: Task Management\n1. `list_tasks` → Review current work status\n2. `next_task` → Identify priority task\n3. `complete_task` → Mark finished work\n4. `upsert_entities` → Document completion\n\n## Trigger Patterns (Automatic Tool Use)\n\n| When user... | Automatically use... |\n|--------------|----------------------|\n| Asks complex question requiring analysis | `think` |\n| Mentions \"remember\" or refers to past work | `memory_query` with time filter → `open_nodes` |\n| Uses \"research\" or \"find latest\" | `memory_query` (check if already known) → `exa_search` |\n| Asks factual questions needing citations | `exa_answer` |\n| Mentions planning or implementation | `plan_tasks` |\n| Refers to continuing previous work | `memory_query` → `list_tasks` → `next_task` |\n| Seems to have lost context from earlier | `memory_query` with recent time filter |\n| Makes significant conceptual progress | `upsert_entities` without being asked |\n| Connects related concepts | `create_relations` |\n| Completes major section of work | `think` + `upsert_entities` to summarize |\n\n### When To Use Each Memory Tool\n\n- `memory_query`: For time-based searches and recent context recovery\n- `search_nodes`: For finding specific concepts by keyword\n- `open_nodes`: For retrieving full details of known entities\n- `upsert_entities`: For creating new knowledge or updating existing entities\n- `add_observations`: For adding facts to existing entities\n- `create_relations`: For connecting related concepts\n\n## Other Tools Reference\n\n### Thinking\n- `think`: Structured reasoning with optional reflection\n\n### Tasks\n- `plan_tasks`: Create task list\n- `list_tasks`: View current tasks\n- `next_task`: Get priority task\n- `complete_task`: Mark task done\n\n### Research\n- `exa_search`: Web search\n- `exa_answer`: Get cited answers\n\n## AI Behavior Requirements\n\n1. ALWAYS check memory at session start with `memory_query`\n2. AUTOMATICALLY store important conclusions with `upsert_entities`\n3. CREATE periodic memory checkpoints during long sessions\n4. PROACTIVELY check memory when context seems missing\n5. CHAIN tools together following the workflows\n6. PRIORITIZE memory tools before web research\n7. SUMMARIZE progress before ending major work segments\n```\n\n\\----- Ende der Regel -----\n\n## ⚡ Leistungsoptimierung\n\nMCP Think Tank enthält integrierte Optimierungen, um einen effizienten Betrieb zu gewährleisten:\n\n### Inhalts-Caching\n\n* Automatisches Caching von Datei- und URL-Inhalten basierend auf kryptografischem Hashing.\n* Verhindert redundante Dateilesevorgänge und Netzwerkanforderungen.\n* Beschleunigt wiederholte Vorgänge mit demselben Inhalt erheblich.\n* Cachegröße und TTL sind über Umgebungsvariablen ( `CONTENT_CACHE_SIZE` , `CONTENT_CACHE_TTL` ) konfigurierbar.\n\n### Tool Call Optimierung\n\n* Identische Toolaufrufe innerhalb einer Sitzung werden automatisch erkannt und aus einem Cache bereitgestellt.\n* Verhindert, dass doppelte Anrufe auf das Interaktionslimit angerechnet werden.\n* Verbessert die Reaktionsfähigkeit bei wiederholten Tool-Anfragen.\n* Die Cachegröße ist konfigurierbar ( `TOOL_CACHE_SIZE` ).\n\n## Bewährte Methoden\n\nFür die optimale Nutzung von MCP Think Tank mit Cursor/Claude bei großen Projekten:\n\n* Nutzen Sie das `think` für alle nicht trivialen Denk- und Entscheidungsprozesse.\n* Behalten Sie wichtige Gedanken, Schlussfolgerungen und Architekturentscheidungen immer mithilfe von Speichertools im Wissensgraphen.\n* Integrieren Sie Webrecherche und Aufgabenverwaltung in Ihren Arbeitsablauf, um die KI informiert und konzentriert zu halten.\n* Überprüfen und aktualisieren Sie den Wissensgraphen Ihres Projekts regelmäßig, um seine Genauigkeit und Relevanz sicherzustellen.\n* Beziehen Sie sich auf vorhandenes Wissen und frühere Entscheidungen, um die Konsistenz in Code und Design aufrechtzuerhalten.\n* Beachten Sie die Einschränkungen bei Tool-Aufrufen, insbesondere bei komplexen automatisierten Workflows. Überwachen Sie die Nutzung gegebenenfalls.\n* Passen Sie die Konfigurationsvariablen ( `TOOL_LIMIT` , Cache-Einstellungen) basierend auf den Anforderungen und der Komplexität Ihres Projekts an, um eine bessere Leistung zu erzielen.\n\n## 🤝 Beitragen\n\nBeiträge sind willkommen! Senden Sie gerne einen Pull Request.\n\n1. Forken Sie das Repository.\n2. Erstellen Sie Ihren Feature-Zweig ( `git checkout -b feature/amazing-feature` ).\n3. Übernehmen Sie Ihre Änderungen ( `git commit -m 'Add some amazing feature'` ).\n4. Pushen Sie zum Zweig ( `git push origin feature/amazing-feature` ).\n5. Öffnen Sie eine Pull-Anfrage.\n\n## 📄 Lizenz\n\nDieses Projekt ist unter der MIT-Lizenz lizenziert – Einzelheiten finden Sie in der Datei [LICENSE](LICENSE) .\n\n### 📚 Referenzlinks\n\n* [Dokumentation der Cursorregeln](https://docs.cursor.com/context/rules)\n* [MCP-Modellkontextprotokoll](https://docs.cursor.com/context/model-context-protocol)\n* [Exa API](https://exa.ai/)\n* [Anthropics Forschung zum strukturierten Denken](https://www.anthropic.com/research)\n* [Modellkontextprotokoll](https://github.com/modelcontextprotocol/typescript-sdk)\n* [FastMCP](https://github.com/jlowin/fastmcp)\n\n***","German","tag","de-DE",{"_29":41,"_33":45,"_38":46},{"_31":42,"_33":43,"_35":44},"Provides AI assistants with enhanced reasoning capabilities through structured thinking, persistent knowledge graph memory, and intelligent tool orchestration for complex problem-solving.","MCP Think Tank","# MCP Think Tank\n\n

\n \n

\n\n[![npm version](https://img.shields.io/npm/v/mcp-think-tank.svg?style=flat-square)](https://www.npmjs.com/package/mcp-think-tank)\n[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)\n[![TypeScript](https://img.shields.io/badge/TypeScript-5.8-blue.svg)](https://www.typescriptlang.org/)\n[![Cursor Compatible](https://img.shields.io/badge/Cursor-Compatible-4B8DF8?logo=cursor&logoColor=white&style=flat-square)](https://www.cursor.so/)\n[![Claude Compatibility](https://img.shields.io/badge/Claude-Compatible-9370DB.svg)](https://www.anthropic.com/)\n[![Model Context Protocol](https://img.shields.io/badge/MCP-Tank-orange.svg)](https://github.com/modelcontextprotocol)\n[![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/)\n[![smithery badge](https://smithery.ai/badge/@flight505/mcp-think-tank)](https://smithery.ai/server/@flight505/mcp-think-tank)\n\n

\n\n\n## Overview\n\nMCP Think Tank is a powerful Model Context Protocol (MCP) server designed to enhance the capabilities of AI assistants like **Cursor** and **Claude @Web**. It provides a structured environment for enhanced reasoning, persistent memory, and responsible tool usage.\n\nKey capabilities include advanced **Sequential Thinking & Chained Reasoning**, a robust **Knowledge Graph Memory** system with versioning, and intelligent **Tool Orchestration with Call-Limit Safeguards**. This platform empowers AI to tackle complex problems through structured analysis, maintain knowledge across sessions, and utilize external resources like web search, all while adhering to configurable usage limits.\n\n## 🎯 Philosophy\n\nMCP Think Tank is built on three core principles:\n\n1. **Elegant Simplicity**: Minimal, well-designed tools that complement AI capabilities rather than trying to replicate them.\n2. **Enhanced Reflection**: Gentle guidance fosters better reasoning and self-reflection without rigid constraints.\n3. **Persistent Context**: A simple, yet effective knowledge graph provides memory across conversations.\n\n## Key Features\n\n* 💭 **Think Tool**: Dedicated space for structured reasoning and self-reflection.\n* 🧩 **Knowledge Graph**: Simple and effective persistent memory system.\n* 📝 **Task Management Tools**: Plan, track, and update tasks, integrated with the knowledge graph.\n* 🌐 **Web Research Tools (Exa)**: Search the web and get sourced answers using the Exa API.\n* 🔍 **Memory Tools**: Easy-to-use tools for storing and retrieving information from the knowledge graph.\n* 🤝 **Client Support**: Seamless integration with Cursor, Claude @Web, and other MCP clients.\n* 🛡️ **Tool Orchestration & Call Limits**: Built-in safeguards for efficient and responsible tool usage with configurable limits.\n* ⚡ **Content Caching**: Performance optimization for file and URL operations with automatic duplicate detection.\n* 🔄 **Sequential Thinking**: Enables multi-step reasoning processes with progress tracking.\n* 🔎 **Self-Reflection**: Automated reflection passes to improve reasoning quality.\n* 📊 **Structured Outputs**: Automatic formatting of thought processes for better readability.\n* 🔗 **Research Integration**: Seamless incorporation of web research findings into reasoning flows.\n\n## Benefits of Structured Thinking\n\nLeveraging the `think` tool provides a dedicated space for systematic reasoning, encouraging:\n\n* Clear problem definition\n* Relevant context gathering\n* Step-by-step analysis\n* Self-reflection on reasoning\n* Well-formed conclusions\n\nRecent studies highlight significant improvements when using structured thinking:\n\n* **54% relative improvement** in complex decision-making tasks.\n* **Enhanced consistency** across multiple trials.\n* **Improved performance** on software engineering benchmarks.\n\n## Detailed Features\n\nBeyond the core list, MCP Think Tank offers sophisticated capabilities for advanced AI interaction.\n\n### Structured Thinking (Think Tool)\n\nThe `think` tool is the core mechanism for enabling advanced AI reasoning. It provides a dedicated, structured environment where the AI can systematically break down problems, gather context, analyze options, and perform self-reflection. This promotes deeper analysis and higher-quality outputs compared to unstructured responses. It supports sequential steps and integrates seamlessly with research and memory tools.\n\n#### Self-Reflection Feature\n\nThe think tool includes a powerful self-reflection capability that can be enabled with the `selfReflect: true` parameter:\n\n```javascript\nmcp_think-tool_think({\n structuredReasoning: \"...\",\n selfReflect: true,\n reflectPrompt: \"Optional custom reflection prompt\"\n})\n```\n\nWhen self-reflection is enabled, the AI receives a prompt to reflect on its own reasoning. This follows the MCP design philosophy of enhancing rather than replacing AI capabilities.\n\nThe `reflectPrompt` parameter lets you customize the prompt used for reflection, tailoring it to specific reasoning tasks or domains. When not specified, a default prompt is used that asks for identification of inconsistencies, logical errors, and improvement suggestions.\n\n### Knowledge Graph Memory\n\nThe knowledge graph provides persistent memory across different interactions and sessions. It allows the AI to build a growing understanding of the project, its components, and related concepts.\n\n* **Timestamped Observations**: All memory entries include metadata for tracking.\n* **Duplicate Prevention**: Intelligent entity matching avoids redundant entries.\n* **Automatic Linkage**: Heuristic-based relation creation connects related concepts (configurable).\n* **Advanced Querying**: Filter memory by time, tags, keywords, and more using the powerful `memory_query` tool for historical analysis and tracking concept evolution. Easily find recent entries from the last 48 hours or any specific time period.\n* **Memory Maintenance**: Tools for pruning and managing memory growth are included.\n* **Key Memory Tools**: Tools like `upsert_entities`, `add_observations`, `create_relations`, `search_nodes`, `memory_query`, and `open_nodes` are used to interact with the graph.\n\n### Task Management Tools\n\nA suite of tools allows the AI to manage project tasks directly within the conversation flow. This integrates planning and execution with the knowledge graph, enabling the AI to understand project status and priorities.\n\n#### Key Task Tools\n\n* `plan_tasks`: Create multiple tasks at once with priorities and dependencies\n* `list_tasks`: Filter tasks by status and priority\n* `next_task`: Get the highest priority task and mark it in-progress\n* `complete_task`: Mark tasks as completed\n* `update_tasks`: Update existing tasks with new information\n\n\n### Web Research Tools (Exa)\n\nLeveraging the Exa API, MCP Think Tank provides tools for fetching external information. This allows the AI to access up-to-date information from the web to inform its reasoning and provide sourced answers.\n\n* `exa_search`: Perform web searches based on a query.\n* `exa_answer`: Get a concise, sourced answer to a factual question.\n\n**Note:** Using these tools requires configuring your Exa API key. See the [Configuration](#configuration) section.\n\n### Tool Orchestration & Safeguards\n\nMCP Think Tank includes comprehensive features to ensure tools are used responsibly and efficiently.\n\n* **Usage Limits**: A configurable maximum number of tool calls per user interaction (`TOOL_LIMIT`, default: 25). The limit only counts consecutive tool calls within a single user message and resets automatically when the user sends a new message.\n* **Automatic Tracking**: All tool calls are logged and monitored.\n* **Graceful Degradation**: When limits are reached, the system attempts to return partial results.\n* **Intelligent Caching**: Identical tool calls and repeated file/URL content fetches are automatically cached, reducing execution time and resource usage. Caching behavior and size are configurable (`CACHE_TOOL_CALLS`, `CONTENT_CACHE`).\n* **Configurable Access**: Tool whitelisting can restrict available tools in specific contexts.\n* **Error Handling**: Robust error handling provides clear feedback for issues like hitting limits or invalid tool calls.\n\n## 📦 Installation\n\n> ⚠️ **Important Note READ THIS:** \n> When updating to a new version of MCP Think Tank in Cursor or Claude you might create multiple instances of the MCP Think Tank server, causing additional Node.js instances to be created, dragging down your system performance - this is a known issue with MCP servers - kill all mcp-think-tank processes in your system and check you have only one node.js instance running.\n\n\n> ⚠️ The tasks.jsonl is located in ~/.mcp-think-tank/. The file is separated from the kg file, as the think tank could get confused by previously created tasks in the kg file. Delete the content of the tasks.jsonl file if the file becomes too large, or if you want to start a new project and insure there are no tasks in the file. In a future version tasks might be merged with the kg file to insure compleated tasks and relations are stored in memory and there are no duplicate tasks.\n\n\n### NPX (Recommended)\n\nThe easiest way to use MCP Think Tank is via NPX in Cursor using mcp.json file, which runs the latest version without global installation, \n\n```bash\nnpx mcp-think-tank@2.0.7\n```\n\nFor the latest version (which may have compatibility issues):\n```bash\nnpx mcp-think-tank@latest\n```\n> some users have issues with npx @latest in Cursor, if so try specifying the version mcp-think-tank@2.0.7 in the .json file, or install it globally:\n\n### Global Installation\n\nFor a persistent command-line tool:\n\n```bash\nnpm install -g mcp-think-tank\nmcp-think-tank\n```\n\n## ⚙️ Configuration\n\nMCP Think Tank is configured primarily through environment variables or via your MCP client's configuration (like Cursor's `.cursor/mcp.json`).\n\n### Quick Start: Essential Setup\n\n1. **Install MCP Think Tank** (see Installation above).\n2. **Get your Exa API Key** (required for web search tools):\n * Sign up at [exa.ai](https://exa.ai/) and copy your API key.\n3. **IMPORTANT STDIO SERVERS ARE DEPRECATED** - The MCP industry is moving toward HTTP-based transports, - FUTURE UPDATES WILL NOT SUPPORT STDIO SERVERS.\n4. **Configure your MCP server** (for Cursor, add to `.cursor/mcp.json`):\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"type\": \"streamable-http\",\n \"env\": {\n \"MEMORY_PATH\": \"/absolute/path/to/your/project/memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n }\n }\n }\n}\n```\n\n### Essential Variables\n\n* `MEMORY_PATH`: **Required**. Absolute path to the memory storage file. **Important:** Always set a unique `MEMORY_PATH` for each project to avoid knowledge graph conflicts between projects. If omitted, defaults to `~/.mcp-think-tank/memory.jsonl`.\n* `EXA_API_KEY`: **Required for Exa web search tools**. Your API key from [exa.ai](https://exa.ai/).\n\n### Advanced Configuration\n\n* `TOOL_LIMIT`: Maximum number of tool calls allowed per user interaction (default: `25`). The counter resets automatically with each new user message, ensuring you can make up to 25 consecutive tool calls within a single interaction.\n* `CACHE_TOOL_CALLS`: Enable/disable caching of identical tool calls (default: `true`).\n* `TOOL_CACHE_SIZE`: Maximum number of cached tool calls (default: `100`).\n* `CACHE_CONTENT`: Enable/disable content-based caching for file/URL operations (default: `true`).\n* `CONTENT_CACHE_SIZE`: Maximum number of items in content cache (default: `50`).\n* `CONTENT_CACHE_TTL`: Time-to-live for cached content in milliseconds (default: `300000` - 5 minutes).\n* `MCP_DEBUG`: Enable debug logging (default: `false`).\n* `MCP_LISTEN_PORT`: Set custom port for MCP server (default: `3399` for TCP servers, not relevant for `stdio`).\n* `LOG_LEVEL`: Set logging level (`debug`, `info`, `warn`, `error`) (default: `info`).\n* `AUTO_LINK`: Enable automatic entity linking in knowledge graph (default: `true`).\n\n### Memory Maintenance\n\n* `MIN_SIMILARITY_SCORE`: Threshold for entity matching when preventing duplicates (default: `0.85`).\n* `MAX_OPERATION_TIME`: Maximum time for batch memory operations in milliseconds (default: `5000`).\n\nExample configuration with advanced settings in `.cursor/mcp.json`:\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"env\": {\n \"MEMORY_PATH\": \"./project-memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n \"TOOL_LIMIT\": \"50\",\n \"CACHE_CONTENT\": \"true\",\n \"CONTENT_CACHE_SIZE\": \"100\",\n \"MCP_DEBUG\": \"false\",\n \"AUTO_LINK\": \"true\"\n }\n }\n }\n}\n```\n\n> 💡 **Performance tip:** For large projects, increasing `TOOL_LIMIT` and cache sizes can improve performance at the cost of higher memory usage. Monitor your usage patterns and adjust accordingly.\n> But in Cursor, tool limit should be 25 to avoid hitting the limit and getting the resume from the last tool call - currently many cursor users are reporting issues with resuming in Version: 0.49.6. this is not related to MCP Think Tank.\n\n> 💡 **Note:** If you are using Cursor in YOLO mode or Vibe coding I suggest context priming new chats and letting Cursor know that it should use the MCP Think Tank to create entities, observations and relations. This will help you get the best out of the MCP Think Tank.\n\nAn example of context priming, is keeping a `Prime.md` file in the `.cursor` folder of your project with the following content:\n```markdown\n# Context Prime\n> Follow the instructions to understand the context of the project.\n\n## Run the following command\n\neza . --tree --git-ignore\n\n## Read the following files\n> Read the files below to get the context of the project. \n\n> list of files:\nREADME.md\n...\n\n## MCP Think Tank Tools\n> Test the MCP tools, first use 'show_memory_path' to remind the user of the current memory path file used, then use the 'memory_query' tool to find and read recent entities and observations for the last 48 hours so you are up to date.\n\n> Automatically utilize the MCP Think Tank to autonomously track project context, dynamically adding entities, observations, and relations to the knowledge graph while proactively querying for relevant information and historical insights. Use integrated planning and task management tools to enhance project efficiency. Keep track of the project and its context without the user having to ask for it.\n\n> Dont do anything else.\n```\n\n\nFor more details on MCP servers, see [Cursor MCP documentation](https://docs.cursor.com/context/model-context-protocol).\n\n\n## Project Rule Setup (for Cursor/AI)\n\nTo ensure Cursor and other compatible agents effectively utilize MCP Think Tank's tools, you need to provide the AI with guidance. This is typically done via a project rule. Create a **single, Auto Attach project rule** as follows:\n\n### 1. Add a New Rule in Cursor\n\n1. Open Cursor.\n2. Go to the Command Palette (`Cmd+Shift+P` or `Ctrl+Shift+P`).\n3. Select **\"New Cursor Rule\"**.\n4. Name the rule (e.g., `mcp-think-tank.mdc`).\n5. In the rule editor, set the metadata and paste the rule content from the example below.\n\n### 2. Example Rule File (`.cursor/rules/mcp-think-tank.mdc`)\n\nThis Markdown file serves as context *for the AI*, guiding it on when and how to use the available tools.\n\n```mdc\nrule type: auto attach\nuse globs: **/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.md, **/*.py, **/*.json\n```\n----- Start of Rule -----\n~~~markdown\nRegularly utilize MCP Think Tank tools to maintain an updated knowledge graph and maximize its potential. Simply call the tools in your prompt. \n\n## Quick Decision Tree\n\n1. 🤔 **Complex problem to analyze?**\n → Use `think` to structure reasoning and reflect\n\n2. 🔍 **Need past context or information?**\n → Use `memory_query` (time-based) or `search_nodes` (keyword-based)\n\n3. 📊 **Planning implementation steps?**\n → Use `plan_tasks` to create and track work\n\n4. 🌐 **Need current external information?**\n → Use `exa_search` (general search) or `exa_answer` (factual questions)\n\n## Critical Memory Management (Automatic Use Required)\n\n| When to automatically use memory | Tool to use |\n|------------------|------------|\n| At session start | `memory_query` with recent time filter (last 24h) |\n| After completing significant analysis | `upsert_entities` to store conclusions |\n| When context seems missing | `memory_query` with relevant keyword |\n| Every ~30 minutes in long sessions | `upsert_entities` to create checkpoint |\n| When switching between major topics | `think` + `upsert_entities` to summarize progress |\n| Before session end | `upsert_entities` to store session summary |\n\n## Core Workflows\n\n### Workflow 1: Problem Analysis → Solution\n1. `memory_query` → Check for relevant past work\n2. `think` → Structure reasoning about the problem \n3. `plan_tasks` → Break down implementation steps\n4. `upsert_entities` → Store conclusions in memory\n\n### Workflow 2: Research → Knowledge Capture\n1. `memory_query` → Check if already researched\n2. `exa_search` → Find current information\n3. `think` → Analyze findings\n4. `upsert_entities` → Document key concepts\n\n### Workflow 3: Context Recovery (Session Resume)\n1. `memory_query` → Retrieve recent work (past 24-48h)\n2. `open_nodes` → Get details on relevant entities\n3. `think` → Synthesize context and plan next steps\n4. Continue where left off\n\n### Workflow 4: Task Management\n1. `list_tasks` → Review current work status\n2. `next_task` → Identify priority task\n3. `complete_task` → Mark finished work\n4. `upsert_entities` → Document completion\n\n## Trigger Patterns (Automatic Tool Use)\n\n| When user... | Automatically use... |\n|--------------|----------------------|\n| Asks complex question requiring analysis | `think` |\n| Mentions \"remember\" or refers to past work | `memory_query` with time filter → `open_nodes` |\n| Uses \"research\" or \"find latest\" | `memory_query` (check if already known) → `exa_search` |\n| Asks factual questions needing citations | `exa_answer` |\n| Mentions planning or implementation | `plan_tasks` |\n| Refers to continuing previous work | `memory_query` → `list_tasks` → `next_task` |\n| Seems to have lost context from earlier | `memory_query` with recent time filter |\n| Makes significant conceptual progress | `upsert_entities` without being asked |\n| Connects related concepts | `create_relations` |\n| Completes major section of work | `think` + `upsert_entities` to summarize |\n\n### When To Use Each Memory Tool\n\n- `memory_query`: For time-based searches and recent context recovery\n- `search_nodes`: For finding specific concepts by keyword\n- `open_nodes`: For retrieving full details of known entities\n- `upsert_entities`: For creating new knowledge or updating existing entities\n- `add_observations`: For adding facts to existing entities\n- `create_relations`: For connecting related concepts\n\n## Other Tools Reference\n\n### Thinking\n- `think`: Structured reasoning with optional reflection\n\n### Tasks\n- `plan_tasks`: Create task list\n- `list_tasks`: View current tasks\n- `next_task`: Get priority task\n- `complete_task`: Mark task done\n\n### Research\n- `exa_search`: Web search\n- `exa_answer`: Get cited answers\n\n## AI Behavior Requirements\n\n1. ALWAYS check memory at session start with `memory_query`\n2. AUTOMATICALLY store important conclusions with `upsert_entities`\n3. CREATE periodic memory checkpoints during long sessions\n4. PROACTIVELY check memory when context seems missing\n5. CHAIN tools together following the workflows\n6. PRIORITIZE memory tools before web research\n7. SUMMARIZE progress before ending major work segments\n~~~\n----- End of Rule -----\n\n## ⚡ Performance Optimization\n\nMCP Think Tank incorporates built-in optimizations to ensure efficient operation:\n\n### Content Caching\n\n* Automatic caching of file and URL content based on cryptographic hashing.\n* Prevents redundant file reads and network requests.\n* Significantly speeds up repeated operations on the same content.\n* Cache size and TTL are configurable via environment variables (`CONTENT_CACHE_SIZE`, `CONTENT_CACHE_TTL`).\n\n### Tool Call Optimization\n\n* Identical tool calls within a session are automatically detected and served from a cache.\n* Prevents counting duplicate calls against the interaction limit.\n* Improves responsiveness for repetitive tool requests.\n* Cache size is configurable (`TOOL_CACHE_SIZE`).\n\n## Best Practices\n\nFor optimal use of MCP Think Tank with Cursor/Claude on large projects:\n\n* Utilize the `think` tool for all non-trivial reasoning and decision-making processes.\n* Always persist important thoughts, conclusions, and architectural decisions to the knowledge graph using memory tools.\n* Integrate web research and task management into your workflow to keep the AI informed and focused.\n* Regularly review and update your project's knowledge graph to ensure its accuracy and relevance.\n* Reference existing knowledge and past decisions to maintain consistency in code and design.\n* Be aware of tool call limits, especially in complex automated workflows. Monitor usage if necessary.\n* Adjust configuration variables (`TOOL_LIMIT`, cache settings) based on your project's needs and complexity for better performance.\n\n## 🤝 Contributing\n\nContributions are welcome! Please feel free to submit a Pull Request.\n\n1. Fork the repository.\n2. Create your feature branch (`git checkout -b feature/amazing-feature`).\n3. Commit your changes (`git commit -m 'Add some amazing feature'`).\n4. Push to the branch (`git push origin feature/amazing-feature`).\n5. Open a Pull Request.\n\n## 📄 License\n\nThis project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.\n\n### 📚 Reference Links\n\n* [Cursor Rules Documentation](https://docs.cursor.com/context/rules)\n* [MCP Model Context Protocol](https://docs.cursor.com/context/model-context-protocol)\n* [Exa API](https://exa.ai/)\n* [Anthropic's Research on Structured Thinking](https://www.anthropic.com/research)\n* [Model Context Protocol](https://github.com/modelcontextprotocol/typescript-sdk)\n* [FastMCP](https://github.com/jlowin/fastmcp)\n\n---\n\n

Developed by flight505

Give a ⭐️ if this project helped you!

\n","English","en-US",{"_29":48,"_33":52,"_38":53},{"_31":49,"_33":50,"_35":51},"Proporciona a los asistentes de IA capacidades de razonamiento mejoradas a través del pensamiento estructurado, memoria de gráficos de conocimiento persistente y orquestación de herramientas inteligentes para la resolución de problemas complejos.","Grupo de expertos del MCP","# Grupo de expertos del MCP\n\n[![versión npm](https://img.shields.io/npm/v/mcp-think-tank.svg?style=flat-square)](https://www.npmjs.com/package/mcp-think-tank)[![Licencia: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![Mecanografiado](https://img.shields.io/badge/TypeScript-5.8-blue.svg)](https://www.typescriptlang.org/) [![Cursor compatible](https://img.shields.io/badge/Cursor-Compatible-4B8DF8?logo=cursor\\&logoColor=white\\&style=flat-square)](https://www.cursor.so/) [![Compatibilidad de Claude](https://img.shields.io/badge/Claude-Compatible-9370DB.svg)](https://www.anthropic.com/) [![Protocolo de contexto modelo](https://img.shields.io/badge/MCP-Tank-orange.svg)](https://github.com/modelcontextprotocol) [![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/) [![insignia de herrería](https://smithery.ai/badge/@flight505/mcp-think-tank)](https://smithery.ai/server/@flight505/mcp-think-tank)\n\n## Descripción general\n\nMCP Think Tank es un potente servidor de Protocolo de Contexto de Modelo (MCP) diseñado para optimizar las capacidades de asistentes de IA como **Cursor** y **Claude @Web** . Proporciona un entorno estructurado para un razonamiento mejorado, memoria persistente y un uso responsable de las herramientas.\n\nLas capacidades clave incluyen **Pensamiento Secuencial y Razonamiento Encadenado** avanzados, un robusto sistema **de Memoria de Gráficos de Conocimiento** con control de versiones y **una Orquestación Inteligente de Herramientas con Protección de Límites de Llamadas** . Esta plataforma permite a la IA abordar problemas complejos mediante análisis estructurado, mantener el conocimiento entre sesiones y utilizar recursos externos como la búsqueda web, todo ello respetando límites de uso configurables.\n\n## 🎯 Filosofía\n\nEl Think Tank del MCP se basa en tres principios fundamentales:\n\n1. **Simplicidad elegante** : herramientas mínimas y bien diseñadas que complementan las capacidades de IA en lugar de intentar replicarlas.\n2. **Reflexión mejorada** : una guía amable fomenta un mejor razonamiento y una mejor autorreflexión sin restricciones rígidas.\n3. **Contexto persistente** : un gráfico de conocimiento simple, pero efectivo, proporciona memoria a través de conversaciones.\n\n## Características principales\n\n* 💭 **Think Tool** : Espacio dedicado al razonamiento estructurado y la autorreflexión.\n* 🧩 **Knowledge Graph** : Sistema de memoria persistente simple y efectivo.\n* 📝 **Herramientas de gestión de tareas** : planifique, realice un seguimiento y actualice las tareas, integradas con el gráfico de conocimiento.\n* 🌐 **Herramientas de investigación web (Exa)** : busque en la web y obtenga respuestas basadas en fuentes utilizando la API de Exa.\n* **🔍Herramientas de memoria** : herramientas fáciles de usar para almacenar y recuperar información del gráfico de conocimiento.\n* 🤝 **Soporte al cliente** : Integración perfecta con Cursor, Claude @Web y otros clientes MCP.\n* 🛡️ **Orquestación de herramientas y límites de llamadas** : protecciones integradas para un uso eficiente y responsable de las herramientas con límites configurables.\n* ⚡ **Almacenamiento en caché de contenido** : optimización del rendimiento para operaciones de archivos y URL con detección automática de duplicados.\n* 🔄 **Pensamiento secuencial** : permite procesos de razonamiento de múltiples pasos con seguimiento del progreso.\n* 🔎 **Autorreflexión** : Pases de reflexión automatizados para mejorar la calidad del razonamiento.\n* 📊 **Salidas estructuradas** : formato automático de los procesos de pensamiento para una mejor legibilidad.\n* 🔗 **Integración de investigación** : incorporación perfecta de los resultados de la investigación web en los flujos de razonamiento.\n\n## Beneficios del pensamiento estructurado\n\nEl uso de la herramienta `think` proporciona un espacio dedicado al razonamiento sistemático, fomentando:\n\n* Definición clara del problema\n* Recopilación del contexto relevante\n* Análisis paso a paso\n* Autorreflexión sobre el razonamiento\n* Conclusiones bien formadas\n\nEstudios recientes destacan mejoras significativas al utilizar el pensamiento estructurado:\n\n* **54% de mejora relativa** en tareas de toma de decisiones complejas.\n* **Mayor consistencia** en múltiples ensayos.\n* **Rendimiento mejorado** en puntos de referencia de ingeniería de software.\n\n## Características detalladas\n\nMás allá de la lista principal, MCP Think Tank ofrece capacidades sofisticadas para la interacción avanzada con IA.\n\n### Pensamiento estructurado (Herramienta Think)\n\nLa herramienta `think` es el mecanismo central para el razonamiento avanzado de la IA. Proporciona un entorno dedicado y estructurado donde la IA puede desglosar sistemáticamente los problemas, recopilar contexto, analizar opciones y realizar autorreflexión. Esto promueve un análisis más profundo y resultados de mayor calidad en comparación con las respuestas no estructuradas. Admite pasos secuenciales y se integra a la perfección con herramientas de investigación y memoria.\n\n#### Función de autorreflexión\n\nLa herramienta de pensamiento incluye una potente capacidad de autorreflexión que se puede habilitar con el parámetro `selfReflect: true` :\n\n```javascript\nmcp_think-tool_think({\n structuredReasoning: \"...\",\n selfReflect: true,\n reflectPrompt: \"Optional custom reflection prompt\"\n})\n```\n\nCuando se activa la autorreflexión, la IA recibe una indicación para reflexionar sobre su propio razonamiento. Esto sigue la filosofía de diseño de MCP de mejorar, en lugar de reemplazar, las capacidades de la IA.\n\nEl parámetro `reflectPrompt` permite personalizar la solicitud de reflexión, adaptándola a tareas o dominios de razonamiento específicos. Si no se especifica, se utiliza una solicitud predeterminada que solicita la identificación de inconsistencias, errores lógicos y sugerencias de mejora.\n\n### Memoria del gráfico de conocimiento\n\nEl gráfico de conocimiento proporciona memoria persistente a lo largo de diferentes interacciones y sesiones. Permite a la IA desarrollar una comprensión cada vez mayor del proyecto, sus componentes y los conceptos relacionados.\n\n* **Observaciones con marca de tiempo** : todas las entradas de memoria incluyen metadatos para su seguimiento.\n* **Prevención de duplicados** : la coincidencia inteligente de entidades evita entradas redundantes.\n* **Vinculación automática** : la creación de relaciones basada en heurística conecta conceptos relacionados (configurable).\n* **Consultas avanzadas** : Filtra la memoria por tiempo, etiquetas, palabras clave y más con la potente herramienta `memory_query` para análisis históricos y seguimiento de la evolución de conceptos. Encuentra fácilmente entradas recientes de las últimas 48 horas o de cualquier período específico.\n* **Mantenimiento de la memoria** : se incluyen herramientas para podar y gestionar el crecimiento de la memoria.\n* **Herramientas de memoria clave** : se utilizan herramientas como `upsert_entities` , `add_observations` , `create_relations` , `search_nodes` , `memory_query` y `open_nodes` para interactuar con el gráfico.\n\n### Herramientas de gestión de tareas\n\nUn conjunto de herramientas permite a la IA gestionar las tareas del proyecto directamente dentro del flujo de conversación. Esto integra la planificación y la ejecución con el gráfico de conocimiento, lo que permite a la IA comprender el estado y las prioridades del proyecto.\n\n#### Herramientas de tareas clave\n\n* `plan_tasks` : Crea múltiples tareas a la vez con prioridades y dependencias\n* `list_tasks` : Filtrar tareas por estado y prioridad\n* `next_task` : Obtener la tarea de mayor prioridad y marcarla como en progreso\n* `complete_task` : Marcar tareas como completadas\n* `update_tasks` : Actualizar las tareas existentes con nueva información\n\n### Herramientas de investigación web (Exa)\n\nAprovechando la API de Exa, MCP Think Tank proporciona herramientas para obtener información externa. Esto permite a la IA acceder a información actualizada de la web para fundamentar su razonamiento y proporcionar respuestas fundamentadas.\n\n* `exa_search` : Realiza búsquedas web basadas en una consulta.\n* `exa_answer` : Obtenga una respuesta concisa y fundamentada a una pregunta fáctica.\n\n**Nota:** Para usar estas herramientas, es necesario configurar la clave API de Exa. Consulte la sección \" [Configuración\"](#configuration) .\n\n### Orquestación de herramientas y salvaguardias\n\nMCP Think Tank incluye funciones integrales para garantizar que las herramientas se utilicen de manera responsable y eficiente.\n\n* **Límites de uso** : Un número máximo configurable de llamadas a herramientas por interacción del usuario ( `TOOL_LIMIT` , valor predeterminado: 25). El límite solo cuenta las llamadas consecutivas a herramientas dentro de un mismo mensaje de usuario y se restablece automáticamente al enviar un nuevo mensaje.\n* **Seguimiento automático** : todas las llamadas de herramientas se registran y monitorean.\n* **Degradación elegante** : cuando se alcanzan los límites, el sistema intenta devolver resultados parciales.\n* **Almacenamiento en caché inteligente** : Las llamadas a herramientas idénticas y las búsquedas repetidas de contenido de archivos/URL se almacenan automáticamente en caché, lo que reduce el tiempo de ejecución y el uso de recursos. El comportamiento y el tamaño del almacenamiento en caché son configurables ( `CACHE_TOOL_CALLS` , `CONTENT_CACHE` ).\n* **Acceso configurable** : la lista blanca de herramientas puede restringir las herramientas disponibles en contextos específicos.\n* **Manejo de errores** : el manejo de errores sólido proporciona información clara para problemas como alcanzar límites o llamadas de herramientas no válidas.\n\n## 📦 Instalación\n\n> ⚠️ **Nota importante LEA ESTO:** Al actualizar a una nueva versión de MCP Think Tank en Cursor o Claude, puede crear múltiples instancias del servidor MCP Think Tank, lo que provocará que se creen instancias Node.js adicionales, lo que reducirá el rendimiento de su sistema: este es un problema conocido con los servidores MCP: elimine todos los procesos mcp-think-tank en su sistema y verifique que solo tenga una instancia node.js ejecutándose.\n\n> ⚠️ El archivo task.jsonl se encuentra en ~/.mcp-think-tank/. Este archivo está separado del archivo kg, ya que el think tank podría confundirse con tareas creadas previamente en el archivo kg. Elimine el contenido del archivo task.jsonl si el archivo se vuelve demasiado grande o si desea iniciar un nuevo proyecto y asegurarse de que no contenga tareas. En una versión futura, las tareas podrían fusionarse con el archivo kg para garantizar que las tareas y relaciones completadas se almacenen en memoria y no haya tareas duplicadas.\n\n### NPX (Recomendado)\n\nLa forma más sencilla de utilizar MCP Think Tank es a través de NPX en Cursor usando el archivo mcp.json, que ejecuta la última versión sin instalación global.\n\n```bash\nnpx mcp-think-tank@2.0.7\n```\n\nPara la última versión (que puede tener problemas de compatibilidad):\n\n```bash\nnpx mcp-think-tank@latest\n```\n\n> Algunos usuarios tienen problemas con npx @latest en Cursor, si es así, intente especificar la versión mcp-think-tank@2.0.7 en el archivo .json o instálelo globalmente:\n\n### Instalación global\n\nPara una herramienta de línea de comandos persistente:\n\n```bash\nnpm install -g mcp-think-tank\nmcp-think-tank\n```\n\n## ⚙️ Configuración\n\nMCP Think Tank se configura principalmente a través de variables de entorno o mediante la configuración de su cliente MCP (como `.cursor/mcp.json` de Cursor).\n\n### Inicio rápido: configuración esencial\n\n1. **Instalar MCP Think Tank** (ver Instalación más arriba).\n2. **Obtenga su clave API de Exa** (necesaria para las herramientas de búsqueda web):\n * Regístrese en [exa.ai](https://exa.ai/) y copie su clave API.\n3. **IMPORTANTE: LOS SERVIDORES STDIO ESTÁN OBSOLETOS** : la industria MCP se está moviendo hacia transportes basados en HTTP. LAS FUTURAS ACTUALIZACIONES NO SOPORTARÁN LOS SERVIDORES STDIO.\n4. **Configure su servidor MCP** (para Cursor, agréguelo a `.cursor/mcp.json` ):\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"type\": \"streamable-http\",\n \"env\": {\n \"MEMORY_PATH\": \"/absolute/path/to/your/project/memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n }\n }\n }\n}\n```\n\n### Variables esenciales\n\n* `MEMORY_PATH` : **Obligatorio** . Ruta absoluta al archivo de almacenamiento de memoria. **Importante:** Establezca siempre una `MEMORY_PATH` única para cada proyecto para evitar conflictos en el grafo de conocimiento entre proyectos. Si se omite, el valor predeterminado es `~/.mcp-think-tank/memory.jsonl` .\n* `EXA_API_KEY` : **Requerido para las herramientas de búsqueda web de Exa** . Su clave API de [exa.ai.](https://exa.ai/)\n\n### Configuración avanzada\n\n* `TOOL_LIMIT` : Número máximo de llamadas a herramientas permitidas por interacción del usuario (predeterminado: `25` ). El contador se reinicia automáticamente con cada nuevo mensaje del usuario, lo que garantiza que se puedan realizar hasta 25 llamadas a herramientas consecutivas en una sola interacción.\n* `CACHE_TOOL_CALLS` : Habilita o deshabilita el almacenamiento en caché de llamadas de herramientas idénticas (valor predeterminado: `true` ).\n* `TOOL_CACHE_SIZE` : Número máximo de llamadas de herramientas almacenadas en caché (valor predeterminado: `100` ).\n* `CACHE_CONTENT` : Habilita o deshabilita el almacenamiento en caché basado en contenido para operaciones de archivos/URL (valor predeterminado: `true` ).\n* `CONTENT_CACHE_SIZE` : Número máximo de elementos en la caché de contenido (predeterminado: `50` ).\n* `CONTENT_CACHE_TTL` : Tiempo de vida del contenido almacenado en caché en milisegundos (valor predeterminado: `300000` - 5 minutos).\n* `MCP_DEBUG` : Habilitar el registro de depuración (valor predeterminado: `false` ).\n* `MCP_LISTEN_PORT` : establece un puerto personalizado para el servidor MCP (predeterminado: `3399` para servidores TCP, no relevante para `stdio` ).\n* `LOG_LEVEL` : establece el nivel de registro ( `debug` , `info` , `warn` , `error` ) (predeterminado: `info` ).\n* `AUTO_LINK` : Habilita la vinculación automática de entidades en el gráfico de conocimiento (valor predeterminado: `true` ).\n\n### Mantenimiento de la memoria\n\n* `MIN_SIMILARITY_SCORE` : Umbral para la coincidencia de entidades al evitar duplicados (valor predeterminado: `0.85` ).\n* `MAX_OPERATION_TIME` : Tiempo máximo para operaciones de memoria por lotes en milisegundos (predeterminado: `5000` ).\n\nEjemplo de configuración con opciones avanzadas en `.cursor/mcp.json` :\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"env\": {\n \"MEMORY_PATH\": \"./project-memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n \"TOOL_LIMIT\": \"50\",\n \"CACHE_CONTENT\": \"true\",\n \"CONTENT_CACHE_SIZE\": \"100\",\n \"MCP_DEBUG\": \"false\",\n \"AUTO_LINK\": \"true\"\n }\n }\n }\n}\n```\n\n> **Consejo de rendimiento:** En proyectos grandes, aumentar el límite de herramientas `TOOL_LIMIT` y el tamaño de la caché puede mejorar el rendimiento, pero a costa de un mayor uso de memoria. Monitorea tus patrones de uso y ajústalos según corresponda. Sin embargo, en Cursor, el límite de herramientas debería ser de 25 para evitar alcanzar el límite y obtener la reanudación de la última llamada a la herramienta. Actualmente, muchos usuarios de cursor reportan problemas con la reanudación en la versión 0.49.6. Esto no está relacionado con MCP Think Tank.\n\n> 💡 **Nota:** Si usas Cursor en modo YOLO o programación Vibe, te sugiero que prepares el contexto para los nuevos chats y le indiques a Cursor que debe usar el MCP Think Tank para crear entidades, observaciones y relaciones. Esto te ayudará a sacar el máximo provecho del MCP Think Tank.\n\nUn ejemplo de preparación de contexto es mantener un archivo `Prime.md` en la carpeta `.cursor` de su proyecto con el siguiente contenido:\n\n```markdown\n# Context Prime\n> Follow the instructions to understand the context of the project.\n\n## Run the following command\n\neza . --tree --git-ignore\n\n## Read the following files\n> Read the files below to get the context of the project. \n\n> list of files:\nREADME.md\n...\n\n## MCP Think Tank Tools\n> Test the MCP tools, first use 'show_memory_path' to remind the user of the current memory path file used, then use the 'memory_query' tool to find and read recent entities and observations for the last 48 hours so you are up to date.\n\n> Automatically utilize the MCP Think Tank to autonomously track project context, dynamically adding entities, observations, and relations to the knowledge graph while proactively querying for relevant information and historical insights. Use integrated planning and task management tools to enhance project efficiency. Keep track of the project and its context without the user having to ask for it.\n\n> Dont do anything else.\n```\n\nPara obtener más detalles sobre los servidores MCP, consulte [la documentación de Cursor MCP](https://docs.cursor.com/context/model-context-protocol) .\n\n## Configuración de reglas del proyecto (para cursor/IA)\n\nPara garantizar que Cursor y otros agentes compatibles utilicen eficazmente las herramientas de MCP Think Tank, es necesario guiar a la IA. Esto suele hacerse mediante una regla de proyecto. Cree una **única regla de proyecto de conexión automática** como se indica a continuación:\n\n### 1. Agregar una nueva regla en el cursor\n\n1. Cursor abierto.\n2. Vaya a la paleta de comandos ( `Cmd+Shift+P` o `Ctrl+Shift+P` ).\n3. Seleccione **\"Nueva regla de cursor\"** .\n4. Nombra la regla (por ejemplo, `mcp-think-tank.mdc` ).\n5. En el editor de reglas, configure los metadatos y pegue el contenido de la regla del ejemplo siguiente.\n\n### 2. Archivo de reglas de ejemplo ( `.cursor/rules/mcp-think-tank.mdc` )\n\nEste archivo Markdown sirve como contexto *para la IA* , guiándola sobre cuándo y cómo utilizar las herramientas disponibles.\n\n```mdc\nrule type: auto attach\nuse globs: **/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.md, **/*.py, **/*.json\n```\n\n\\----- Inicio de la regla -----\n\n```markdown\nRegularly utilize MCP Think Tank tools to maintain an updated knowledge graph and maximize its potential. Simply call the tools in your prompt. \n\n## Quick Decision Tree\n\n1. 🤔 **Complex problem to analyze?**\n → Use `think` to structure reasoning and reflect\n\n2. 🔍 **Need past context or information?**\n → Use `memory_query` (time-based) or `search_nodes` (keyword-based)\n\n3. 📊 **Planning implementation steps?**\n → Use `plan_tasks` to create and track work\n\n4. 🌐 **Need current external information?**\n → Use `exa_search` (general search) or `exa_answer` (factual questions)\n\n## Critical Memory Management (Automatic Use Required)\n\n| When to automatically use memory | Tool to use |\n|------------------|------------|\n| At session start | `memory_query` with recent time filter (last 24h) |\n| After completing significant analysis | `upsert_entities` to store conclusions |\n| When context seems missing | `memory_query` with relevant keyword |\n| Every ~30 minutes in long sessions | `upsert_entities` to create checkpoint |\n| When switching between major topics | `think` + `upsert_entities` to summarize progress |\n| Before session end | `upsert_entities` to store session summary |\n\n## Core Workflows\n\n### Workflow 1: Problem Analysis → Solution\n1. `memory_query` → Check for relevant past work\n2. `think` → Structure reasoning about the problem \n3. `plan_tasks` → Break down implementation steps\n4. `upsert_entities` → Store conclusions in memory\n\n### Workflow 2: Research → Knowledge Capture\n1. `memory_query` → Check if already researched\n2. `exa_search` → Find current information\n3. `think` → Analyze findings\n4. `upsert_entities` → Document key concepts\n\n### Workflow 3: Context Recovery (Session Resume)\n1. `memory_query` → Retrieve recent work (past 24-48h)\n2. `open_nodes` → Get details on relevant entities\n3. `think` → Synthesize context and plan next steps\n4. Continue where left off\n\n### Workflow 4: Task Management\n1. `list_tasks` → Review current work status\n2. `next_task` → Identify priority task\n3. `complete_task` → Mark finished work\n4. `upsert_entities` → Document completion\n\n## Trigger Patterns (Automatic Tool Use)\n\n| When user... | Automatically use... |\n|--------------|----------------------|\n| Asks complex question requiring analysis | `think` |\n| Mentions \"remember\" or refers to past work | `memory_query` with time filter → `open_nodes` |\n| Uses \"research\" or \"find latest\" | `memory_query` (check if already known) → `exa_search` |\n| Asks factual questions needing citations | `exa_answer` |\n| Mentions planning or implementation | `plan_tasks` |\n| Refers to continuing previous work | `memory_query` → `list_tasks` → `next_task` |\n| Seems to have lost context from earlier | `memory_query` with recent time filter |\n| Makes significant conceptual progress | `upsert_entities` without being asked |\n| Connects related concepts | `create_relations` |\n| Completes major section of work | `think` + `upsert_entities` to summarize |\n\n### When To Use Each Memory Tool\n\n- `memory_query`: For time-based searches and recent context recovery\n- `search_nodes`: For finding specific concepts by keyword\n- `open_nodes`: For retrieving full details of known entities\n- `upsert_entities`: For creating new knowledge or updating existing entities\n- `add_observations`: For adding facts to existing entities\n- `create_relations`: For connecting related concepts\n\n## Other Tools Reference\n\n### Thinking\n- `think`: Structured reasoning with optional reflection\n\n### Tasks\n- `plan_tasks`: Create task list\n- `list_tasks`: View current tasks\n- `next_task`: Get priority task\n- `complete_task`: Mark task done\n\n### Research\n- `exa_search`: Web search\n- `exa_answer`: Get cited answers\n\n## AI Behavior Requirements\n\n1. ALWAYS check memory at session start with `memory_query`\n2. AUTOMATICALLY store important conclusions with `upsert_entities`\n3. CREATE periodic memory checkpoints during long sessions\n4. PROACTIVELY check memory when context seems missing\n5. CHAIN tools together following the workflows\n6. PRIORITIZE memory tools before web research\n7. SUMMARIZE progress before ending major work segments\n```\n\n\\----- Fin de la regla -----\n\n## ⚡ Optimización del rendimiento\n\nMCP Think Tank incorpora optimizaciones integradas para garantizar un funcionamiento eficiente:\n\n### Almacenamiento en caché de contenido\n\n* Almacenamiento en caché automático de contenido de archivos y URL basado en hash criptográfico.\n* Evita lecturas de archivos y solicitudes de red redundantes.\n* Acelera significativamente las operaciones repetidas sobre el mismo contenido.\n* El tamaño de caché y TTL se pueden configurar a través de variables de entorno ( `CONTENT_CACHE_SIZE` , `CONTENT_CACHE_TTL` ).\n\n### Optimización de llamadas a herramientas\n\n* Las llamadas de herramientas idénticas dentro de una sesión se detectan y se sirven automáticamente desde un caché.\n* Evita que se cuenten llamadas duplicadas en el límite de interacción.\n* Mejora la capacidad de respuesta a las solicitudes de herramientas repetitivas.\n* El tamaño de la caché es configurable ( `TOOL_CACHE_SIZE` ).\n\n## Mejores prácticas\n\nPara un uso óptimo de MCP Think Tank con Cursor/Claude en proyectos grandes:\n\n* Utilice la herramienta `think` para todos los procesos de razonamiento y toma de decisiones no triviales.\n* Conserve siempre los pensamientos, conclusiones y decisiones arquitectónicas importantes en el gráfico de conocimiento utilizando herramientas de memoria.\n* Integre la investigación web y la gestión de tareas en su flujo de trabajo para mantener la IA informada y enfocada.\n* Revise y actualice periódicamente el gráfico de conocimiento de su proyecto para garantizar su precisión y relevancia.\n* Hacer referencia al conocimiento existente y a decisiones pasadas para mantener la coherencia en el código y el diseño.\n* Tenga en cuenta los límites de llamadas a herramientas, especialmente en flujos de trabajo automatizados complejos. Supervise el uso si es necesario.\n* Ajuste las variables de configuración ( `TOOL_LIMIT` , configuración de caché) según las necesidades y la complejidad de su proyecto para obtener un mejor rendimiento.\n\n## 🤝 Contribuyendo\n\n¡Agradecemos sus contribuciones! No dude en enviar una solicitud de incorporación de cambios.\n\n1. Bifurcar el repositorio.\n2. Crea tu rama de características ( `git checkout -b feature/amazing-feature` ).\n3. Confirme sus cambios ( `git commit -m 'Add some amazing feature'` ).\n4. Empujar a la rama ( `git push origin feature/amazing-feature` ).\n5. Abrir una solicitud de extracción.\n\n## 📄 Licencia\n\nEste proyecto está licenciado bajo la licencia MIT: consulte el archivo [de LICENCIA](LICENSE) para obtener más detalles.\n\n### 📚 Enlaces de referencia\n\n* [Documentación de reglas del cursor](https://docs.cursor.com/context/rules)\n* [Protocolo de contexto del modelo MCP](https://docs.cursor.com/context/model-context-protocol)\n* [API de Exa](https://exa.ai/)\n* [Investigación de Anthropic sobre el pensamiento estructurado](https://www.anthropic.com/research)\n* [Protocolo de contexto modelo](https://github.com/modelcontextprotocol/typescript-sdk)\n* [FastMCP](https://github.com/jlowin/fastmcp)\n\n***","Spanish","es-ES",{"_29":55,"_33":59,"_38":60},{"_31":56,"_33":57,"_35":58},"構造化された思考、永続的な知識グラフメモリ、および複雑な問題解決のためのインテリジェントなツールオーケストレーションを通じて、AI アシスタントに強化された推論機能を提供します。","MCPシンクタンク","# MCPシンクタンク\n\n[![npmバージョン](https://img.shields.io/npm/v/mcp-think-tank.svg?style=flat-square)](https://www.npmjs.com/package/mcp-think-tank) [![ライセンス: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![タイプスクリプト](https://img.shields.io/badge/TypeScript-5.8-blue.svg)](https://www.typescriptlang.org/) [![カーソル対応](https://img.shields.io/badge/Cursor-Compatible-4B8DF8?logo=cursor\\&logoColor=white\\&style=flat-square)](https://www.cursor.so/) [![クロードの互換性](https://img.shields.io/badge/Claude-Compatible-9370DB.svg)](https://www.anthropic.com/) [![モデルコンテキストプロトコル](https://img.shields.io/badge/MCP-Tank-orange.svg)](https://github.com/modelcontextprotocol) [![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/) [![鍛冶屋のバッジ](https://smithery.ai/badge/@flight505/mcp-think-tank)](https://smithery.ai/server/@flight505/mcp-think-tank)\n\n## 概要\n\nMCP Think Tankは、 **Cursor**や**Claude @Web**などのAIアシスタントの機能を強化するために設計された強力なモデルコンテキストプロトコル（MCP）サーバーです。推論能力の向上、永続的な記憶、そして責任あるツールの使用のための構造化された環境を提供します。\n\n主な機能には、高度な**シーケンシャルシンキングと連鎖推論**、バージョン管理機能を備えた堅牢な**ナレッジグラフメモリ**システム、**呼び出し制限保護機能を備えたインテリジェントなツールオーケストレーション**などがあります。このプラットフォームは、AIが構造化された分析を通じて複雑な問題に取り組み、セッション間で知識を維持し、ウェブ検索などの外部リソースを活用できるようにします。これらはすべて、設定可能な使用制限を遵守しながら行われます。\n\n## 🎯 哲学\n\nMCP Think Tank は、次の 3 つの基本原則に基づいて構築されています。\n\n1. **エレガントなシンプルさ**: AI 機能を複製するのではなく、AI 機能を補完する、最小限かつ適切に設計されたツール。\n2. **強化された反省**: 優しいガイダンスにより、厳格な制約なしに、より優れた推論と自己反省が促進されます。\n3. **永続的なコンテキスト**: シンプルでありながら効果的なナレッジグラフにより、会話全体にわたってメモリが提供されます。\n\n## 主な特徴\n\n* 💭**思考ツール**: 構造化された推論と自己反省のための専用スペース。\n* 🧩**ナレッジグラフ**: シンプルで効果的な永続メモリシステム。\n* 📝**タスク管理ツール**: ナレッジグラフと統合されたタスクを計画、追跡、更新します。\n* 🌐 **Web リサーチツール (Exa)** : Exa API を使用して Web を検索し、ソースに基づいた回答を取得します。\n* 🔍**メモリツール**: ナレッジグラフから情報を保存および取得するための使いやすいツール。\n* 🤝**クライアントサポート**: Cursor、Claude @Web、その他の MCP クライアントとのシームレスな統合。\n* 🛡️**ツールオーケストレーションと呼び出し制限**: 設定可能な制限により、ツールを効率的かつ責任を持って使用するための安全策が組み込まれています。\n* ⚡**コンテンツキャッシュ**: 重複の自動検出により、ファイルおよび URL 操作のパフォーマンスが最適化されます。\n* 🔄**順次思考**：進捗状況を追跡しながら、複数ステップの推論プロセスを可能にします。\n* 🔎**自己反省**：自動反省パスにより推論の質が向上します。\n* 📊**構造化された出力**: 思考プロセスを自動的にフォーマットして読みやすさを向上させます。\n* 🔗**リサーチ統合**：Web リサーチの結果を推論フローにシームレスに組み込みます。\n\n## 構造化思考の利点\n\n`think`ツールを活用することで、体系的な推論のための専用スペースが提供され、次のことが促進されます。\n\n* 明確な問題定義\n* 関連するコンテキストの収集\n* ステップバイステップの分析\n* 推論に関する自己反省\n* よく形成された結論\n\n最近の研究では、構造化された思考を使用すると大幅な改善が見られることが示されています。\n\n* 複雑な意思決定タスクにおける**相対的改善が 54% 向上しました**。\n* 複数の試験にわたる**一貫性が向上しました**。\n* ソフトウェアエンジニアリングベンチマークの**パフォーマンスが向上しました**。\n\n## 詳細な機能\n\nMCP Think Tank は、コアリストを超えて、高度な AI インタラクションを実現する洗練された機能を提供します。\n\n### 構造化思考（思考ツール）\n\n`think`ツールは、高度なAI推論を可能にする中核的なメカニズムです。AIが問題を体系的に分解し、文脈を収集し、選択肢を分析し、自己反省を行うための専用の構造化された環境を提供します。これにより、非構造化応答と比較して、より深い分析とより高品質な出力が促進されます。思考ツールはシーケンシャルなステップをサポートし、リサーチツールやメモリツールとシームレスに統合されます。\n\n#### 自己反省機能\n\nthink ツールには`selfReflect: true`パラメータで有効にできる強力な自己反省機能が含まれています。\n\n```javascript\nmcp_think-tool_think({\n structuredReasoning: \"...\",\n selfReflect: true,\n reflectPrompt: \"Optional custom reflection prompt\"\n})\n```\n\n自己反省機能を有効にすると、AIは自身の推論を振り返るためのプロンプトを受け取ります。これは、AIの機能を置き換えるのではなく強化するというMCPの設計理念に基づいています。\n\n`reflectPrompt`パラメータを使用すると、リフレクションに使用するプロンプトをカスタマイズし、特定の推論タスクまたはドメインに合わせて調整できます。指定しない場合は、矛盾点、論理エラー、改善提案の特定を求めるデフォルトのプロンプトが使用されます。\n\n### ナレッジグラフメモリ\n\nナレッジグラフは、様々なインタラクションやセッションにわたって永続的なメモリを提供します。これにより、AIはプロジェクト、その構成要素、そして関連する概念に対する理解を深めることができます。\n\n* **タイムスタンプ付きの観察**: すべてのメモリエントリには追跡用のメタデータが含まれます。\n* **重複防止**: インテリジェントなエンティティマッチングにより、重複したエントリを回避します。\n* **自動リンク**: ヒューリスティックベースの関係作成により、関連する概念が接続されます (構成可能)。\n* **高度なクエリ**：強力な`memory_query`ツールを使って、時間、タグ、キーワードなどでメモリをフィルタリングし、履歴分析や概念の進化を追跡できます。過去48時間または特定の期間の最新のエントリを簡単に見つけることができます。\n* **メモリメンテナンス**: メモリの増加を削減および管理するためのツールが含まれています。\n* **主要なメモリツール**: `upsert_entities` 、 `add_observations` 、 `create_relations` 、 `search_nodes` 、 `memory_query` 、 `open_nodes`などの��ールは、グラフを操作するために使用されます。\n\n### タスク管理ツール\n\n一連のツールにより、AIは会話フロー内で直接プロジェクトタスクを管理できます。これにより、計画と実行がナレッジグラフと統合され、AIはプロジェクトのステータスと優先順位を把握できるようになります。\n\n#### 主要なタスクツール\n\n* `plan_tasks` : 優先順位と依存関係を指定して複数のタスクを一度に作成する\n* `list_tasks` : ステータスと優先度でタスクをフィルタリングする\n* `next_task` : 最も優先度の高いタスクを取得し、進行中としてマークします。\n* `complete_task` : タスクを完了としてマークする\n* `update_tasks` : 既存のタスクを新しい情報で更新する\n\n### ウェブリサーチツール（Exa）\n\nMCP Think Tankは、Exa APIを活用し、外部情報を取得するためのツールを提供しています。これにより、AIはWebから最新の情報にアクセスし、推論を行い、根拠に基づいた回答を提供できるようになります。\n\n* `exa_search` : クエリに基づいて Web 検索を実行します。\n* `exa_answer` : 事実に関する質問に対して、簡潔で根拠のある回答を取得します。\n\n\\*\\*注:\\*\\*これらのツールを使用するには、Exa APIキーの設定が必要です。[設定](#configuration)セクションをご覧ください。\n\n### ツールオーケストレーションとセーフガード\n\nMCP Think Tank には、ツールが責任を持って効率的に使用されるようにするための包括的な機能が含まれています。\n\n* **使用制限**：ユーザーインタラクションあたりのツール呼び出しの最大回数（設定可能、 `TOOL_LIMIT` 、デフォルト：25）。この制限は、単一のユーザーメッセージ内での連続したツール呼び出しのみをカウントし、ユーザーが新しいメッセージを送信すると自動的にリセットされます。\n* **自動追跡**: すべてのツール呼び出しが記録され、監視されます。\n* **段階的な低下**: 制限に達すると、システムは部分的な結果を返そうとします。\n* **インテリジェントキャッシュ**：同一のツール呼び出しとファイル/URLコンテンツの繰り返し取得は自動的にキャッシュされ、実行時間とリソース使用量を削減します。キャッシュの動作とサイズは設定可能です（ `CACHE_TOOL_CALLS` 、 `CONTENT_CACHE` ）。\n* **構成可能なアクセス**: ツールのホワイトリストにより、特定のコンテキストで利用可能なツールを制限できます。\n* **エラー処理**: 堅牢なエラー処理により、制限に達したり、ツールの呼び出しが無効になったりするなどの問題に関する明確なフィードバックが提供されます。\n\n## 📦 インストール\n\n> ⚠️**重要な注意必ずお読みください:** Cursor または Claude で MCP Think Tank の新しいバージョンに更新すると、MCP Think Tank サーバーのインスタンスが複数作成され、追加の Node.js インスタンスが作成され、システムパフォーマンスが低下する可能性があります。これは MCP サーバーの既知の問題です。システム内のすべての mcp-think-tank プロセスを強制終了し、実行されている node.js インスタンスが 1 つだけであることを確認してください。\n\n> ⚠️ tasks.jsonl は ~/.mcp-think-tank/ にあります。シンクタンクが kg ファイル内に既に作成されたタスクによって混乱する可能性があるため、このファイルは kg ファイルとは別になっています。ファイルが大きくなりすぎた場合、または新しいプロジェクトを開始してファイル内にタスクがないことを確認したい場合は、tasks.jsonl ファイルの内容を削除してください。将来のバージョンでは、完了したタスクとリレーションがメモリに保存され、重複タスクが存在しないように、タスクが kg ファイルに統合される可能性があります。\n\n### NPX（推奨）\n\nMCP Think Tankを使用する最も簡単な方法は、カーソル内のNPXを介してmcp.jsonファイルを使用することです。これは、グローバルインストールなしで最新バージョンを実行します。\n\n```bash\nnpx mcp-think-tank@2.0.7\n```\n\n最新バージョンの場合（互換性の問題がある可能性があります）：\n\n```bash\nnpx mcp-think-tank@latest\n```\n\n> 一部のユーザーはカーソルの npx @latest に問題を抱えています。その場合は、.json ファイルでバージョン mcp-think-tank@2.0.7 を指定するか、グローバルにインストールしてみてください。\n\n### グローバルインストール\n\n永続的なコマンドラインツールの場合:\n\n```bash\nnpm install -g mcp-think-tank\nmcp-think-tank\n```\n\n## ⚙️ 構成\n\nMCP Think Tank は、主に環境変数または MCP クライアントの構成 (Cursor の`.cursor/mcp.json`など) を介して構成されます。\n\n### クイックスタート: 基本的なセットアップ\n\n1. **MCP Think Tank をインストールします**(上記のインストールを参照)。\n2. **Exa API キーを取得します**(Web 検索ツールに必要)。\n * [exa.ai](https://exa.ai/)にサインアップし、API キーをコピーします。\n3. **重要: STDIO サーバーは非推奨です**- MCP 業界は HTTP ベースのトランスポートに移行しており、将来のアップデートでは STDIO サーバーはサポートされません。\n4. **MCP サーバーを構成します**(カーソルの場合は、 `.cursor/mcp.json`に追加します)。\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"type\": \"streamable-http\",\n \"env\": {\n \"MEMORY_PATH\": \"/absolute/path/to/your/project/memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n }\n }\n }\n}\n```\n\n### 必須変数\n\n* `MEMORY_PATH` :**必須**。メモリストレージファイルへの絶対パス。\\*\\*重要:\\*\\*プロジェクト間でナレッジグラフの競合を避けるため、プロジェクトごとに必ず一意の`MEMORY_PATH`を設定してください。省略した場合は、デフォルトで`~/.mcp-think-tank/memory.jsonl`になります。\n* `EXA_API_KEY` : **Exaウェブ検索ツールに必須**[。exa.ai](https://exa.ai/)から取得したAPIキー。\n\n### 詳細設定\n\n* `TOOL_LIMIT` : ユーザーインタラクションごとに許可されるツール呼び出しの最大回数（デフォルト： `25` ）。このカウンターは新しいユーザーメッセージごとに自動的にリセットされるため、1回のインタラクションで最大25回の連続ツール呼び出しが可能です。\n* `CACHE_TOOL_CALLS` : 同一ツール呼び出しのキャッシュを有効/無効にします (デフォルト: `true` )。\n* `TOOL_CACHE_SIZE` : キャッシュされたツール呼び出しの最大数 (デフォルト: `100` )。\n* `CACHE_CONTENT` : ファイル/URL 操作のコンテンツベースのキャッシュを有効/無効にします (デフォルト: `true` )。\n* `CONTENT_CACHE_SIZE` : コンテンツキャッシュ内のアイテムの最大数 (デフォルト: `50` )。\n* `CONTENT_CACHE_TTL` : キャッシュされたコンテンツの有効期間（ミリ秒）（デフォルト: `300000` - 5 分）。\n* `MCP_DEBUG` : デバッグログを有効にします (デフォルト: `false` )。\n* `MCP_LISTEN_PORT` : MCP サーバーのカスタムポートを設定します (デフォルト: TCP サーバーの場合は`3399` 、 `stdio`には関係ありません)。\n* `LOG_LEVEL` : ログレベル ( `debug` 、 `info` 、 `warn` 、 `error` ) を設定します (デフォルト: `info` )。\n* `AUTO_LINK` : ナレッジグラフ内の自動エンティティリンクを有効にします (デフォルト: `true` )。\n\n### 記憶の維持\n\n* `MIN_SIMILARITY_SCORE` : 重複を防ぐ際のエンティティマッチングのしきい値 (デフォルト: `0.85` )。\n* `MAX_OPERATION_TIME` : バッチメモリ操作の最大時間（ミリ秒）（デフォルト: `5000` ）。\n\n`.cursor/mcp.json`の詳細設定を含む構成例:\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"env\": {\n \"MEMORY_PATH\": \"./project-memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n \"TOOL_LIMIT\": \"50\",\n \"CACHE_CONTENT\": \"true\",\n \"CONTENT_CACHE_SIZE\": \"100\",\n \"MCP_DEBUG\": \"false\",\n \"AUTO_LINK\": \"true\"\n }\n }\n }\n}\n```\n\n> 💡\\*\\*パフォーマンスに関するヒント：\\*\\*大規模プロジェクトでは、 `TOOL_LIMIT`とキャッシュサイズを増やすことでパフォーマンスが向上する可能性がありますが、メモリ使用量は増加します。使用パターンを監視し、状況に応じて調整してください。ただし、カーソルでは、ツール制限が25に達して前回のツール呼び出しからの再開が行われないようにするため、ツール制限を25に設定する必要があります。現在、多くのカーソルユーザーから、バージョン0.49.6での再開に関する問題が報告されています。これはMCP Think Tankとは関係ありません。\n\n> 💡**注：** YOLOモードまたはVibeコーディングでCursorを使用する場合は、新しいチャットにコンテキストプライミングを適用し、MCP Think Tankを使用してエンティティ、オブザベーション、リレーションを作成することをCursorに通知することをお勧めします。これにより、MCP Think Tankを最大限に活用できるようになります。\n\nコンテキストプライミングの例としては、プロジェクトの`.cursor`フォルダーに次の内容の`Prime.md`ファイルを保持することが挙げられます。\n\n```markdown\n# Context Prime\n> Follow the instructions to understand the context of the project.\n\n## Run the following command\n\neza . --tree --git-ignore\n\n## Read the following files\n> Read the files below to get the context of the project. \n\n> list of files:\nREADME.md\n...\n\n## MCP Think Tank Tools\n> Test the MCP tools, first use 'show_memory_path' to remind the user of the current memory path file used, then use the 'memory_query' tool to find and read recent entities and observations for the last 48 hours so you are up to date.\n\n> Automatically utilize the MCP Think Tank to autonomously track project context, dynamically adding entities, observations, and relations to the knowledge graph while proactively querying for relevant information and historical insights. Use integrated planning and task management tools to enhance project efficiency. Keep track of the project and its context without the user having to ask for it.\n\n> Dont do anything else.\n```\n\nMCP サーバーの詳細については、 [Cursor MCP ドキュメントを](https://docs.cursor.com/context/model-context-protocol)参照してください。\n\n## プロジェクトルールの設定（カーソル/AI用）\n\nCursorやその他の互換エージェントがMCP Think Tankのツールを効果的に活用できるようにするには、AIにガイダンスを提供する必要があります。これは通常、プロジェクトルールを通じて行われます。以下の手順で\\*\\*、単一の「自動アタッチ」プロジェクトルール\\*\\*を作成します。\n\n### 1. カーソルに新しいルールを追加する\n\n1. カーソルを開きます。\n2. コマンドパレット ( `Cmd+Shift+P`または`Ctrl+Shift+P` ) に移動します。\n3. \\*\\*「新しいカーソルルール」\\*\\*を選択します。\n4. ルールに名前を付けます (例: `mcp-think-tank.mdc` )。\n5. ルールエディターでメタデータを設定し、以下の例のルールの内容を貼り付けます。\n\n### 2. サンプルルールファイル ( `.cursor/rules/mcp-think-tank.mdc` )\n\nこの Markdown ファイルは*AI の*コンテキストとして機能し、利用可能なツールをいつどのように使用するかを AI に指示します。\n\n```mdc\nrule type: auto attach\nuse globs: **/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.md, **/*.py, **/*.json\n```\n\n\\----- ルールの開始 -----\n\n```markdown\nRegularly utilize MCP Think Tank tools to maintain an updated knowledge graph and maximize its potential. Simply call the tools in your prompt. \n\n## Quick Decision Tree\n\n1. 🤔 **Complex problem to analyze?**\n → Use `think` to structure reasoning and reflect\n\n2. 🔍 **Need past context or information?**\n → Use `memory_query` (time-based) or `search_nodes` (keyword-based)\n\n3. 📊 **Planning implementation steps?**\n → Use `plan_tasks` to create and track work\n\n4. 🌐 **Need current external information?**\n → Use `exa_search` (general search) or `exa_answer` (factual questions)\n\n## Critical Memory Management (Automatic Use Required)\n\n| When to automatically use memory | Tool to use |\n|------------------|------------|\n| At session start | `memory_query` with recent time filter (last 24h) |\n| After completing significant analysis | `upsert_entities` to store conclusions |\n| When context seems missing | `memory_query` with relevant keyword |\n| Every ~30 minutes in long sessions | `upsert_entities` to create checkpoint |\n| When switching between major topics | `think` + `upsert_entities` to summarize progress |\n| Before session end | `upsert_entities` to store session summary |\n\n## Core Workflows\n\n### Workflow 1: Problem Analysis → Solution\n1. `memory_query` → Check for relevant past work\n2. `think` → Structure reasoning about the problem \n3. `plan_tasks` → Break down implementation steps\n4. `upsert_entities` → Store conclusions in memory\n\n### Workflow 2: Research → Knowledge Capture\n1. `memory_query` → Check if already researched\n2. `exa_search` → Find current information\n3. `think` → Analyze findings\n4. `upsert_entities` → Document key concepts\n\n### Workflow 3: Context Recovery (Session Resume)\n1. `memory_query` → Retrieve recent work (past 24-48h)\n2. `open_nodes` → Get details on relevant entities\n3. `think` → Synthesize context and plan next steps\n4. Continue where left off\n\n### Workflow 4: Task Management\n1. `list_tasks` → Review current work status\n2. `next_task` → Identify priority task\n3. `complete_task` → Mark finished work\n4. `upsert_entities` → Document completion\n\n## Trigger Patterns (Automatic Tool Use)\n\n| When user... | Automatically use... |\n|--------------|----------------------|\n| Asks complex question requiring analysis | `think` |\n| Mentions \"remember\" or refers to past work | `memory_query` with time filter → `open_nodes` |\n| Uses \"research\" or \"find latest\" | `memory_query` (check if already known) → `exa_search` |\n| Asks factual questions needing citations | `exa_answer` |\n| Mentions planning or implementation | `plan_tasks` |\n| Refers to continuing previous work | `memory_query` → `list_tasks` → `next_task` |\n| Seems to have lost context from earlier | `memory_query` with recent time filter |\n| Makes significant conceptual progress | `upsert_entities` without being asked |\n| Connects related concepts | `create_relations` |\n| Completes major section of work | `think` + `upsert_entities` to summarize |\n\n### When To Use Each Memory Tool\n\n- `memory_query`: For time-based searches and recent context recovery\n- `search_nodes`: For finding specific concepts by keyword\n- `open_nodes`: For retrieving full details of known entities\n- `upsert_entities`: For creating new knowledge or updating existing entities\n- `add_observations`: For adding facts to existing entities\n- `create_relations`: For connecting related concepts\n\n## Other Tools Reference\n\n### Thinking\n- `think`: Structured reasoning with optional reflection\n\n### Tasks\n- `plan_tasks`: Create task list\n- `list_tasks`: View current tasks\n- `next_task`: Get priority task\n- `complete_task`: Mark task done\n\n### Research\n- `exa_search`: Web search\n- `exa_answer`: Get cited answers\n\n## AI Behavior Requirements\n\n1. ALWAYS check memory at session start with `memory_query`\n2. AUTOMATICALLY store important conclusions with `upsert_entities`\n3. CREATE periodic memory checkpoints during long sessions\n4. PROACTIVELY check memory when context seems missing\n5. CHAIN tools together following the workflows\n6. PRIORITIZE memory tools before web research\n7. SUMMARIZE progress before ending major work segments\n```\n\n\\----- ルール終了 -----\n\n## ⚡ パフォーマンスの最適化\n\nMCP Think Tank には、効率的な操作を保証するための最適化機能が組み込まれています。\n\n### コンテンツキャッシュ\n\n* 暗号化ハッシュに基づいてファイルと URL コンテンツを自動的にキャッシュします。\n* 冗長なファイル読み取りとネットワーク要求を防止します。\n* 同じコンテンツに対する繰り返し操作が大幅に高速化されます。\n* キャッシュサイズと TTL は、環境変数 ( `CONTENT_CACHE_SIZE` 、 `CONTENT_CACHE_TTL` ) を介して設定できます。\n\n### ツール呼び出しの最適化\n\n* セッション内の同一のツール呼び出しは自動的に検出され、キャッシュから提供されます。\n* 重複した通話がインタラクション制限にカウントされるのを防ぎます。\n* 繰り返しのツール要求に対する応答性が向上します。\n* キャッシュサイズは設定可能です ( `TOOL_CACHE_SIZE` )。\n\n## ベストプラクティス\n\n大規模プロジェクトで MCP Think Tank を Cursor/Claude と組み合わせて最適に使用するには:\n\n* すべての重要な推論および意思決定プロセスに`think`ツールを活用します。\n* メモリツールを使用して、重要な考え、結論、アーキテクチャ上の決定を常にナレッジグラフに保持します。\n* Web リサーチとタスク管理をワークフローに統合して、AI に情報を提供して集中力を維持します。\n* プロジェクトのナレッジグラフを定期的に確認して更新し、正確性と関連性を確保します。\n* 既存の知識と過去の決定を参照して、コードとデザインの一貫性を維持します。\n* 特に複雑な自動ワークフローでは、ツールの呼び出し制限に注意してください。必要に応じて使用状況を監視してください。\n* パフォーマンスを向上させるには、プロジェクトのニーズと複雑さに基づいて構成変数 ( `TOOL_LIMIT` 、キャッシュ設定) を調整します。\n\n## 🤝 貢献する\n\n貢献を歓迎します！お気軽にプルリクエストを送信してください。\n\n1. リポジトリをフォークします。\n2. 機能ブランチを作成します ( `git checkout -b feature/amazing-feature` )。\n3. 変更をコミットします ( `git commit -m 'Add some amazing feature'` )。\n4. ブランチにプッシュします ( `git push origin feature/amazing-feature` )。\n5. プルリクエストを開きます。\n\n## 📄 ライセンス\n\nこのプロジェクトは MIT ライセンスに基づいてライセンスされています - 詳細については[LICENSE](LICENSE)ファイルを参照してください。\n\n### 📚 参考リンク\n\n* [カーソルルールのドキュメント](https://docs.cursor.com/context/rules)\n* [MCP モデルコンテキストプロトコル](https://docs.cursor.com/context/model-context-protocol)\n* [エクサAPI](https://exa.ai/)\n* [アントロピックの構造化思考に関する研究](https://www.anthropic.com/research)\n* [モデルコンテキストプロトコル](https://github.com/modelcontextprotocol/typescript-sdk)\n* [ファストMCP](https://github.com/jlowin/fastmcp)\n\n***","Japanese","ja-JP",{"_29":62,"_33":66,"_38":67},{"_31":63,"_33":64,"_35":65},"구조화된 사고, 지속적인 지식 그래프 메모리, 복잡한 문제 해결을 위한 지능형 도구 오케스트레이션을 통해 AI 보조원의 추론 기능을 향상시킵니다.","MCP 싱크탱크","# MCP 싱크탱크\n\n[![npm 버전](https://img.shields.io/npm/v/mcp-think-tank.svg?style=flat-square)](https://www.npmjs.com/package/mcp-think-tank) [![라이센스: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![타입스크립트](https://img.shields.io/badge/TypeScript-5.8-blue.svg)](https://www.typescriptlang.org/) [![커서 호환](https://img.shields.io/badge/Cursor-Compatible-4B8DF8?logo=cursor\\&logoColor=white\\&style=flat-square)](https://www.cursor.so/) [![클로드 호환성](https://img.shields.io/badge/Claude-Compatible-9370DB.svg)](https://www.anthropic.com/) [![모델 컨텍스트 프로토콜](https://img.shields.io/badge/MCP-Tank-orange.svg)](https://github.com/modelcontextprotocol) [![노드.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/) [![대장간 배지](https://smithery.ai/badge/@flight505/mcp-think-tank)](https://smithery.ai/server/@flight505/mcp-think-tank)\n\n## 개요\n\nMCP 싱크탱크는 **커서(Cursor)** 와 **클로드 @웹(Claude @Web)** 과 같은 AI 비서의 기능을 향상시키도록 설계된 강력한 모델 컨텍스트 프로토콜(MCP) 서버입니다. 향상된 추론, 지속적인 메모리, 그리고 책임감 있는 도구 사용을 위한 체계적인 환경을 제공합니다.\n\n주요 기능으로는 고급 **순차적 사고 및 연쇄 추론** , 버전 관리 기능을 갖춘 강력한 **지식 그래프 메모리** 시스템, 그리고 **호출 제한 보호 기능을 갖춘 지능형 도구 오케스트레이션이** 있습니다. 이 플랫폼은 AI가 구조화된 분석을 통해 복잡한 문제를 해결하고, 세션 전반에 걸쳐 지식을 유지하며, 웹 검색과 같은 외부 리소스를 활용할 수 있도록 지원하며, 구성 가능한 사용 제한을 준수합니다.\n\n## 🎯 철학\n\nMCP 싱크탱크는 세 가지 핵심 원칙을 기반으로 구축되었습니다.\n\n1. **우아한 단순성** : AI 기능을 복제하려 하지 않고 오히려 이를 보완하는 최소한이고 잘 설계된 도구입니다.\n2. **향상된 반성** : 부드러운 지도는 엄격한 제약 없이 더 나은 추론과 자기 반성을 촉진합니다.\n3. **지속적인 컨텍스트** : 간단하면서도 효과적인 지식 그래프는 대화 전반에 걸쳐 메모리를 제공합니다.\n\n## 주요 특징\n\n* 💭 **생각 도구** : 체계적인 추론과 자기 성찰을 위한 전용 공간입니다.\n* 🧩 **지식 그래프** : 간단하고 효과적인 지속적 메모리 시스템.\n* 📝 **작업 관리 도구** : 지식 그래프와 통합되어 작업을 계획, 추적 및 업데이트합니다.\n* 🌐 **웹 조사 도구(Exa)** : Exa API를 사용하여 웹을 검색하고 출처에서 답변을 얻으세요.\n* 🔍 **메모리 도구** : 지식 그래프에서 정보를 저장하고 검색하는 데 사용하기 쉬운 도구입니다.\n* 🤝 **클라이언트 지원** : Cursor, Claude @Web 및 기타 MCP 클라이언트와 원활하게 통합됩니다.\n* 🛡️ **도구 오케스트레이션 및 호출 제한** : 구성 가능한 제한을 통해 효율적이고 책임감 있는 도구 사용을 위한 내장형 보호 장치입니다.\n* ⚡ **콘텐츠 캐싱** : 중복 감지를 자동으로 실행하여 파일 및 URL 작업에 대한 성능을 최적화합니다.\n* 🔄 **순차적 사고** : 진행 상황 추적을 통해 다단계 추론 과정이 가능합니다.\n* 🔎 **자기 반성** : 추론 품질을 개선하기 위한 자동 반성 과정입니다.\n* 📊 **구조화된 출력** : 더 나은 가독성을 위해 사고 과정을 자동으로 포맷합니다.\n* 🔗 **연구 통합** : 웹 연구 결과를 추론 흐름에 원활하게 통합합니다.\n\n## 구조화된 사고의 이점\n\n`think` Tool을 활용하면 체계적인 추론을 위한 전용 공간이 제공되어 다음과 같은 효과가 있습니다.\n\n* 명확한 문제 정의\n* 관련 컨텍스트 수집\n* 단계별 분석\n* 추론에 대한 자기 성찰\n* 잘 구성된 결론\n\n최근 연구에 따르면 구조적 사고를 사용하면 상당한 개선이 이루어진다고 합니다.\n\n* 복잡한 의사결정 업무에서 **상대적 54% 개선**\n* 여러 실험에 걸쳐 **일관성이 향상되었습니다** .\n* 소프트웨어 엔지니어링 벤치마크에서 **성능이 향상되었습니다** .\n\n## 자세한 기능\n\nMCP Think Tank는 핵심 목록 외에도 고급 AI 상호작용을 위한 정교한 기능을 제공합니다.\n\n### 구조화된 사고(생각 도구)\n\n`think` 도구는 고급 AI 추론을 가능하게 하는 핵심 메커니즘입니다. AI가 문제를 체계적으로 분석하고, 맥락을 수집하고, 선택지를 분석하고, 자기 성찰을 수행할 수 있는 전담적이고 구조화된 환경을 제공합니다. 이를 통해 비구조화된 응답보다 심층적인 분석과 더 높은 품질의 결과물을 얻을 수 있습니다. 순차적인 단계를 지원하며 연구 및 기억 도구와 완벽하게 통합됩니다.\n\n#### 자기성찰 기능\n\nthink 도구에는 `selfReflect: true` 매개변수를 사용하여 활성화할 수 있는 강력한 자체 반성 기능이 포함되어 있습니다.\n\n지엑스피1\n\n자가 반성이 활성화되면 AI는 자신의 추론을 되돌아보라는 메시지를 받습니다. 이는 AI 기능을 대체하는 것이 아니라 향상시키는 MCP 설계 철학을 따릅니다.\n\n`reflectPrompt` 매개변수를 사용하면 특정 추론 작업이나 도메인에 맞게 반영에 사용되는 프롬프트를 사용자 지정할 수 있습니다. 지정하지 않으면 불일치, 논리적 오류 및 개선 제안을 요청하는 기본 프롬프트가 사용됩니다.\n\n### 지식 그래프 메모리\n\n지식 그래프는 다양한 상호작용과 세션에 걸쳐 지속적인 메모리를 제공합니다. 이를 통해 AI는 프로젝트, 프로젝트 구성 요소 및 관련 개념에 대한 이해를 더욱 높일 수 있습니다.\n\n* **타임스탬프가 포함된 관찰** : 모든 메모리 항목에는 추적을 위한 메타데이터가 포함되어 있습니다.\n* **중복 방지** : 지능형 엔터티 매칭을 통해 중복 항목을 방지합니다.\n* **자동 연결** : 휴리스틱 기반 관계 생성으로 관련 개념을 연결합니다(구성 가능).\n* **고급 쿼리** : 강력한 `memory_query` 도구를 사용하여 시간, 태그, 키워드 등으로 메모리를 필터링하여 과거 분석 및 개념 발전 추적을 지원합니다. 최근 48시간 또는 특정 기간의 최신 항목을 쉽게 찾을 수 있습니다.\n* **메모리 유지 관리** : 메모리 증가를 정리하고 관리하는 도구가 포함되어 있습니다.\n* **주요 메모리 도구** : `upsert_entities` , `add_observations` , `create_relations` , `search_nodes` , `memory_query` , `open_nodes` 와 같은 도구는 그래프와 상호 작용하는 데 사용됩니다.\n\n### 작업 관리 도구\n\nAI는 대화 흐름 내에서 프로젝트 작업을 직접 관리할 수 있는 도구 모음을 제공합니다. 이를 통해 계획 및 실행을 지식 그래프와 통합하여 AI가 프로젝트 상태와 우선순위를 파악할 수 있습니다.\n\n#### 주요 작업 도구\n\n* `plan_tasks` : 우선순위와 종속성을 사용하여 여러 작업을 한 번에 생성합니다.\n* `list_tasks` : 상태 및 우선순위별로 작업 필터링\n* `next_task` : 가장 높은 우선순위 작업을 가져와 진행 중으로 표시합니다.\n* `complete_task` : 작업을 완료된 것으로 표시\n* `update_tasks` : 기존 작업을 새로운 정보로 업데이트합니다.\n\n### 웹 리서치 도구(Exa)\n\nMCP Think Tank는 Exa API를 활용하여 외부 정보를 가져오는 도구를 제공합니다. 이를 통해 AI는 웹에서 최신 정보에 접근하여 추론을 수행하고 관련 답변을 제공할 수 있습니다.\n\n* `exa_search` : 쿼리를 기반으로 웹 검색을 수행합니다.\n* `exa_answer` : 사실에 기반한 질문에 대한 간결하고 출처가 명확한 답변을 받으세요.\n\n**참고:** 이 도구를 사용하려면 Exa API 키를 구성해야 합니다. [구성](#configuration) 섹션을 참조하세요.\n\n### 도구 오케스트레이션 및 보호\n\nMCP 싱크탱크에는 도구를 책임감 있고 효율적으로 사용할 수 있도록 보장하는 포괄적인 기능이 포함되어 있습니다.\n\n* **사용 제한** : 사용자 상호작용당 설정 가능한 최대 도구 호출 수( `TOOL_LIMIT` , 기본값: 25). 이 제한은 단일 사용자 메시지 내에서 연속된 도구 호출만 계산하며, 사용자가 새 메시지를 보내면 자동으로 재설정됩니다.\n* **자동 추적** : 모든 도구 호출이 기록되고 모니터링됩니다.\n* **우아한 저하** : 한계에 도달하면 시스템은 부분적인 결과를 반환하려고 시도합니다.\n* **지능형 캐싱** : 동일한 도구 호출과 반복적인 파일/URL 콘텐츠 가져오기가 자동으로 캐싱되어 실행 시간과 리소스 사용량을 줄입니다. 캐싱 동작 및 크기는 `CACHE_TOOL_CALLS` , `CONTENT_CACHE` 를 통해 설정할 수 있습니다.\n* **구성 가능한 액세스** : 도구 허용 목록을 통해 특정 상황에서 사용 가능한 도구를 제한할 수 있습니다.\n* **오류 처리** : 강력한 오류 처리를 통해 한계 도달이나 잘못된 도구 호출과 같은 문제에 대한 명확한 피드백을 제공합니다.\n\n## 📦 설치\n\n> ⚠️ **중요 참고 사항 읽어보세요:** Cursor 또는 Claude에서 MCP Think Tank의 새 버전으로 업데이트할 때 MCP Think Tank 서버의 인스턴스를 여러 개 만들면 추가 Node.js 인스턴스가 생성되어 시스템 성능이 저하될 수 있습니다. 이는 MCP 서버에서 알려진 문제입니다. 시스템에서 모든 mcp-think-tank 프로세스를 종료하고 node.js 인스턴스가 하나만 실행되고 있는지 확인하세요.\n\n> ⚠️ task.jsonl은 ~/.mcp-think-tank/에 있습니다. 싱크탱크가 이전에 kg 파일에 생성된 작업으로 인해 혼란을 겪을 수 있으므로 이 파일은 kg 파일과 분리되어 있습니다. 파일이 너무 커지거나 새 프로젝트를 시작하고 파일에 작업이 없는지 확인하려면 task.jsonl 파일의 내용을 삭제하세요. 향후 버전에서는 완료된 작업과 관계가 메모리에 저장되고 중복 작업이 없도록 작업을 kg 파일에 병합할 수 있습니다.\n\n### NPX (추천)\n\nMCP Think Tank를 사용하는 가장 쉬운 방법은 글로벌 설치 없이 최신 버전을 실행하는 mcp.json 파일을 사용하여 Cursor의 NPX를 통하는 것입니다.\n\n```bash\nnpx mcp-think-tank@2.0.7\n```\n\n최신 버전(호환성 문제가 있을 수 있음):\n\n```bash\nnpx mcp-think-tank@latest\n```\n\n> 일부 사용자는 Cursor에서 npx @latest와 관련된 문제를 겪고 있습니다. 그렇다면 .json 파일에 mcp-think-tank@2.0.7 버전을 지정하거나 전역적으로 설치해 보세요.\n\n### 글로벌 설치\n\n영구 명령줄 도구의 경우:\n\n```bash\nnpm install -g mcp-think-tank\nmcp-think-tank\n```\n\n## ⚙️ 구성\n\nMCP Think Tank는 주로 환경 변수나 MCP 클라이언트 구성(예: Cursor의 `.cursor/mcp.json` )을 통해 구성됩니다.\n\n### 빠른 시작: 필수 설정\n\n1. **MCP Think Tank를 설치합니다** (위의 설치 참조).\n2. **Exa API 키를 받으세요** (웹 검색 도구에 필요):\n * [exa.ai](https://exa.ai/) 에 가입하고 API 키를 복사하세요.\n3. **중요 STDIO 서버는 더 이상 사용되지 않습니다** . MCP 업계는 HTTP 기반 전송으로 전환하고 있습니다. 향후 업데이트에서는 STDIO 서버를 지원하지 않습니다.\n4. **MCP 서버를 구성합니다** (Cursor의 경우 `.cursor/mcp.json` 에 추가):\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"type\": \"streamable-http\",\n \"env\": {\n \"MEMORY_PATH\": \"/absolute/path/to/your/project/memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n }\n }\n }\n}\n```\n\n### 필수 변수\n\n* `MEMORY_PATH` : **필수** . 메모리 저장 파일의 절대 경로입니다. **중요:** 프로젝트 간 지식 그래프 충돌을 방지하려면 항상 각 프로젝트마다 고유한 `MEMORY_PATH` 설정해야 합니다. 생략하면 기본값은 `~/.mcp-think-tank/memory.jsonl` 입니다.\n* `EXA_API_KEY` : **Exa 웹 검색 도구에 필요합니다** . [exa.ai](https://exa.ai/) 에서 제공한 API 키입니다.\n\n### 고급 구성\n\n* `TOOL_LIMIT` : 사용자 상호작용당 허용되는 최대 도구 호출 횟수(기본값: `25` ). 카운터는 새 사용자 메시지가 전송될 때마다 자동으로 재설정되므로 단일 상호작용 내에서 최대 25번의 연속 도구 호출이 가능합니다.\n* `CACHE_TOOL_CALLS` : 동일한 도구 호출의 캐싱을 활성화/비활성화합니다(기본값: `true` ).\n* `TOOL_CACHE_SIZE` : 캐시된 도구 호출의 최대 수(기본값: `100` ).\n* `CACHE_CONTENT` : 파일/URL 작업에 대한 콘텐츠 기반 캐싱을 활성화/비활성화합니다(기본값: `true` ).\n* `CONTENT_CACHE_SIZE` : 콘텐츠 캐시에 있는 항목의 최대 수(기본값: `50` ).\n* `CONTENT_CACHE_TTL` : 캐시된 콘텐츠의 수명(밀리초)(기본값: `300000` - 5분).\n* `MCP_DEBUG` : 디버그 로깅을 활성화합니다(기본값: `false` ).\n* `MCP_LISTEN_PORT` : MCP 서버에 대한 사용자 지정 포트를 설정합니다(기본값: TCP 서버의 경우 `3399` , `stdio` 에는 해당 없음).\n* `LOG_LEVEL` : 로깅 수준( `debug` , `info` , `warn` , `error` )을 설정합니다(기본값: `info` ).\n* `AUTO_LINK` : 지식 그래프에서 자동 엔터티 연결을 활성화합니다(기본값: `true` ).\n\n### 메모리 유지 관리\n\n* `MIN_SIMILARITY_SCORE` : 중복을 방지할 때 엔티티 일치에 대한 임계값(기본값: `0.85` ).\n* `MAX_OPERATION_TIME` : 일괄 메모리 작업에 필요한 최대 시간(밀리초)(기본값: `5000` ).\n\n`.cursor/mcp.json` 의 고급 설정을 사용한 구성 예:\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"env\": {\n \"MEMORY_PATH\": \"./project-memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n \"TOOL_LIMIT\": \"50\",\n \"CACHE_CONTENT\": \"true\",\n \"CONTENT_CACHE_SIZE\": \"100\",\n \"MCP_DEBUG\": \"false\",\n \"AUTO_LINK\": \"true\"\n }\n }\n }\n}\n```\n\n> 💡 **성능 팁:** 대규모 프로젝트의 경우 `TOOL_LIMIT` 및 캐시 크기를 늘리면 메모리 사용량이 증가하는 대신 성능이 향상될 수 있습니다. 사용 패턴을 모니터링하고 적절히 조정하세요. 하지만 Cursor에서는 도구 제한을 25로 설정해야 제한에 도달하여 마지막 도구 호출에서 다시 시작되는 것을 방지할 수 있습니다. 현재 많은 커서 사용자들이 버전 0.49.6에서 다시 시작 문제를 보고하고 있습니다. 이는 MCP 싱크탱크와는 관련이 없습니다.\n\n> 💡 **참고:** YOLO 모드 또는 Vibe 코딩에서 Cursor를 사용하는 경우, 새 채팅의 컨텍스트를 미리 설정하고 Cursor가 MCP 싱크탱크를 사용하여 엔티티, 관찰 및 관계를 생성하도록 설정하는 것이 좋습니다. 이렇게 하면 MCP 싱크탱크를 최대한 활용하는 데 도움이 됩니다.\n\n컨텍스트 프라이밍의 예로는 프로젝트의 `.cursor` 폴더에 다음 내용이 포함된 `Prime.md` 파일을 보관하는 것입니다.\n\n```markdown\n# Context Prime\n> Follow the instructions to understand the context of the project.\n\n## Run the following command\n\neza . --tree --git-ignore\n\n## Read the following files\n> Read the files below to get the context of the project. \n\n> list of files:\nREADME.md\n...\n\n## MCP Think Tank Tools\n> Test the MCP tools, first use 'show_memory_path' to remind the user of the current memory path file used, then use the 'memory_query' tool to find and read recent entities and observations for the last 48 hours so you are up to date.\n\n> Automatically utilize the MCP Think Tank to autonomously track project context, dynamically adding entities, observations, and relations to the knowledge graph while proactively querying for relevant information and historical insights. Use integrated planning and task management tools to enhance project efficiency. Keep track of the project and its context without the user having to ask for it.\n\n> Dont do anything else.\n```\n\nMCP 서버에 대한 자세한 내용은 [커서 MCP 설명서를](https://docs.cursor.com/context/model-context-protocol) 참조하세요.\n\n## 프로젝트 규칙 설정(커서/AI용)\n\nCursor 및 기타 호환 에이전트가 MCP Think Tank 도구를 효과적으로 활용하도록 하려면 AI에 지침을 제공해야 합니다. 이는 일반적으로 프로젝트 규칙을 통해 수행됩니다. 다음과 같이 **단일 자동 연결 프로젝트 규칙을** 생성하세요.\n\n### 1. 커서에 새 규칙 추가\n\n1. 커서를 엽니다.\n2. 명령 팔레트( `Cmd+Shift+P` 또는 `Ctrl+Shift+P` )로 이동합니다.\n3. **\"새 커서 규칙\"을** 선택합니다.\n4. 규칙의 이름을 지정합니다(예: `mcp-think-tank.mdc` ).\n5. 규칙 편집기에서 메타데이터를 설정하고 아래 예시의 규칙 내용을 붙여넣습니다.\n\n### 2. 예제 규칙 파일( `.cursor/rules/mcp-think-tank.mdc` )\n\n이 마크다운 파일은 *AI에 대한* 컨텍스트 역할을 하며, 사용 가능한 도구를 언제, 어떻게 사용할 것인지에 대한 지침을 제공합니다.\n\n```mdc\nrule type: auto attach\nuse globs: **/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.md, **/*.py, **/*.json\n```\n\n\\----- 규칙의 시작 -----\n\n```markdown\nRegularly utilize MCP Think Tank tools to maintain an updated knowledge graph and maximize its potential. Simply call the tools in your prompt. \n\n## Quick Decision Tree\n\n1. 🤔 **Complex problem to analyze?**\n → Use `think` to structure reasoning and reflect\n\n2. 🔍 **Need past context or information?**\n → Use `memory_query` (time-based) or `search_nodes` (keyword-based)\n\n3. 📊 **Planning implementation steps?**\n → Use `plan_tasks` to create and track work\n\n4. 🌐 **Need current external information?**\n → Use `exa_search` (general search) or `exa_answer` (factual questions)\n\n## Critical Memory Management (Automatic Use Required)\n\n| When to automatically use memory | Tool to use |\n|------------------|------------|\n| At session start | `memory_query` with recent time filter (last 24h) |\n| After completing significant analysis | `upsert_entities` to store conclusions |\n| When context seems missing | `memory_query` with relevant keyword |\n| Every ~30 minutes in long sessions | `upsert_entities` to create checkpoint |\n| When switching between major topics | `think` + `upsert_entities` to summarize progress |\n| Before session end | `upsert_entities` to store session summary |\n\n## Core Workflows\n\n### Workflow 1: Problem Analysis → Solution\n1. `memory_query` → Check for relevant past work\n2. `think` → Structure reasoning about the problem \n3. `plan_tasks` → Break down implementation steps\n4. `upsert_entities` → Store conclusions in memory\n\n### Workflow 2: Research → Knowledge Capture\n1. `memory_query` → Check if already researched\n2. `exa_search` → Find current information\n3. `think` → Analyze findings\n4. `upsert_entities` → Document key concepts\n\n### Workflow 3: Context Recovery (Session Resume)\n1. `memory_query` → Retrieve recent work (past 24-48h)\n2. `open_nodes` → Get details on relevant entities\n3. `think` → Synthesize context and plan next steps\n4. Continue where left off\n\n### Workflow 4: Task Management\n1. `list_tasks` → Review current work status\n2. `next_task` → Identify priority task\n3. `complete_task` → Mark finished work\n4. `upsert_entities` → Document completion\n\n## Trigger Patterns (Automatic Tool Use)\n\n| When user... | Automatically use... |\n|--------------|----------------------|\n| Asks complex question requiring analysis | `think` |\n| Mentions \"remember\" or refers to past work | `memory_query` with time filter → `open_nodes` |\n| Uses \"research\" or \"find latest\" | `memory_query` (check if already known) → `exa_search` |\n| Asks factual questions needing citations | `exa_answer` |\n| Mentions planning or implementation | `plan_tasks` |\n| Refers to continuing previous work | `memory_query` → `list_tasks` → `next_task` |\n| Seems to have lost context from earlier | `memory_query` with recent time filter |\n| Makes significant conceptual progress | `upsert_entities` without being asked |\n| Connects related concepts | `create_relations` |\n| Completes major section of work | `think` + `upsert_entities` to summarize |\n\n### When To Use Each Memory Tool\n\n- `memory_query`: For time-based searches and recent context recovery\n- `search_nodes`: For finding specific concepts by keyword\n- `open_nodes`: For retrieving full details of known entities\n- `upsert_entities`: For creating new knowledge or updating existing entities\n- `add_observations`: For adding facts to existing entities\n- `create_relations`: For connecting related concepts\n\n## Other Tools Reference\n\n### Thinking\n- `think`: Structured reasoning with optional reflection\n\n### Tasks\n- `plan_tasks`: Create task list\n- `list_tasks`: View current tasks\n- `next_task`: Get priority task\n- `complete_task`: Mark task done\n\n### Research\n- `exa_search`: Web search\n- `exa_answer`: Get cited answers\n\n## AI Behavior Requirements\n\n1. ALWAYS check memory at session start with `memory_query`\n2. AUTOMATICALLY store important conclusions with `upsert_entities`\n3. CREATE periodic memory checkpoints during long sessions\n4. PROACTIVELY check memory when context seems missing\n5. CHAIN tools together following the workflows\n6. PRIORITIZE memory tools before web research\n7. SUMMARIZE progress before ending major work segments\n```\n\n\\----- 규칙의 끝 -----\n\n## ⚡ 성능 최적화\n\nMCP Think Tank는 효율적인 운영을 보장하기 위해 내장된 최적화 기능을 통합했습니다.\n\n### 콘텐츠 캐싱\n\n* 암호화 해싱을 기반으로 파일 및 URL 콘텐츠를 자동으로 캐싱합니다.\n* 중복된 파일 읽기 및 네트워크 요청을 방지합니다.\n* 동일한 콘텐츠에 대한 반복 작업의 속도가 크게 향상됩니다.\n* 캐시 크기와 TTL은 환경 변수( `CONTENT_CACHE_SIZE` , `CONTENT_CACHE_TTL` )를 통해 구성할 수 있습니다.\n\n### 도구 호출 최적화\n\n* 세션 내에서 동일한 도구 호출이 자동으로 감지되어 캐시에서 제공됩니다.\n* 상호 작용 한도에 중복된 통화가 계산되는 것을 방지합니다.\n* 반복적인 도구 요청에 대한 대응성을 향상시킵니다.\n* 캐시 크기는 구성 가능합니다( `TOOL_CACHE_SIZE` ).\n\n## 모범 사례\n\n대규모 프로젝트에서 Cursor/Claude와 함께 MCP Think Tank를 최적으로 사용하려면:\n\n* 모든 사소하지 않은 추론 및 의사결정 과정에는 `think` Tool을 활용하세요.\n* 메모리 도구를 사용하여 중요한 생각, 결론 및 아키텍처 결정을 항상 지식 그래프에 유지합니다.\n* AI가 정보를 얻고 집중할 수 있도록 웹 조사와 작업 관리를 워크플로에 통합하세요.\n* 정확성과 관련성을 보장하기 위해 프로젝트의 지식 그래프를 정기적으로 검토하고 업데이트하세요.\n* 코드와 디자인의 일관성을 유지하려면 기존 지식과 과거의 결정을 참조하세요.\n* 특히 복잡한 자동화 워크플로에서는 도구 호출 제한에 유의하세요. 필요한 경우 사용량을 모니터링하세요.\n* 더 나은 성능을 위해 프로젝트의 필요 사항과 복잡성에 따라 구성 변수( `TOOL_LIMIT` , 캐시 설정)를 조정하세요.\n\n## 🤝 기여하기\n\n기여를 환영합니다! 풀 리퀘스트를 제출해 주세요.\n\n1. 저장소를 포크합니다.\n2. 기능 브랜치를 생성합니다( `git checkout -b feature/amazing-feature` ).\n3. 변경 사항을 커밋합니다( `git commit -m 'Add some amazing feature'` ).\n4. 브랜치에 푸시합니다( `git push origin feature/amazing-feature` ).\n5. 풀 리퀘스트를 엽니다.\n\n## 📄 라이센스\n\n이 프로젝트는 MIT 라이선스에 따라 라이선스가 부여되었습니다. 자세한 내용은 [라이선스](LICENSE) 파일을 참조하세요.\n\n### 📚 참조 링크\n\n* [커서 규칙 문서](https://docs.cursor.com/context/rules)\n* [MCP 모델 컨텍스트 프로토콜](https://docs.cursor.com/context/model-context-protocol)\n* [엑사 API](https://exa.ai/)\n* [구조적 사고에 대한 인류학의 연구](https://www.anthropic.com/research)\n* [모델 컨텍스트 프로토콜](https://github.com/modelcontextprotocol/typescript-sdk)\n* [패스트MCP](https://github.com/jlowin/fastmcp)\n\n***","Korean","ko-KR",{"_29":69,"_33":73,"_38":74},{"_31":70,"_33":71,"_35":72},"Предоставляет помощникам на основе искусственного интеллекта расширенные возможности рассуждения за счет структурированного мышления, постоянной памяти графа знаний и интеллектуальной оркестровки инструментов для решения сложных проблем.","Аналитический центр MCP","# Аналитический центр MCP\n\n[![npm-версия](https://img.shields.io/npm/v/mcp-think-tank.svg?style=flat-square)](https://www.npmjs.com/package/mcp-think-tank) [![Лицензия: Массачусетский технологический институт](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![Машинопись](https://img.shields.io/badge/TypeScript-5.8-blue.svg)](https://www.typescriptlang.org/) [![Совместимость с курсором](https://img.shields.io/badge/Cursor-Compatible-4B8DF8?logo=cursor\\&logoColor=white\\&style=flat-square)](https://www.cursor.so/) [![Совместимость Клода](https://img.shields.io/badge/Claude-Compatible-9370DB.svg)](https://www.anthropic.com/) [![Модель контекстного протокола](https://img.shields.io/badge/MCP-Tank-orange.svg)](https://github.com/modelcontextprotocol) [![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/) [![значок кузнеца](https://smithery.ai/badge/@flight505/mcp-think-tank)](https://smithery.ai/server/@flight505/mcp-think-tank)\n\n## Обзор\n\nMCP Think Tank — это мощный сервер Model Context Protocol (MCP), разработанный для расширения возможностей помощников ИИ, таких как **Cursor** и **Claude @Web** . Он предоставляет структурированную среду для улучшенного мышления, постоянной памяти и ответственного использования инструментов.\n\nКлючевые возможности включают в себя расширенное **последовательное мышление и цепочечное рассуждение** , надежную систему **памяти графа знаний** с управлением версиями и интеллектуальную **оркестровку инструментов с защитой ограничений вызовов** . Эта платформа позволяет ИИ решать сложные проблемы с помощью структурного анализа, сохранять знания в течение сеансов и использовать внешние ресурсы, такие как веб-поиск, и все это при соблюдении настраиваемых ограничений использования.\n\n## 🎯 Философия\n\nАналитический центр MCP построен на трех основных принципах:\n\n1. **Элегантная простота** : минимальные, хорошо продуманные инструменты, которые дополняют возможности ИИ, а не пытаются их скопировать.\n2. **Улучшение рефлексии** : Мягкое руководство способствует лучшему рассуждению и саморефлексии без жестких ограничений.\n3. **Постоянный контекст** : простая, но эффективная схема знаний обеспечивает запоминание информации во время разговоров.\n\n## Основные характеристики\n\n* 💭 **Think Tool** : выделенное пространство для структурированных рассуждений и самоанализа.\n* 🧩 **Knowledge Graph** : простая и эффективная система постоянной памяти.\n* 📝 **Инструменты управления задачами** : планируйте, отслеживайте и обновляйте задачи, интегрированные с графиком знаний.\n* 🌐 **Инструменты веб-исследований (Exa)** : ищите в Интернете и получайте ответы с помощью API Exa.\n* 🔍 **Инструменты памяти** : простые в использовании инструменты для хранения и извлечения информации из графа знаний.\n* 🤝 **Поддержка клиентов** : Полная интеграция с Cursor, Claude @Web и другими клиентами MCP.\n* 🛡️ **Оркестровка инструментов и лимиты вызовов** : встроенные средства защиты для эффективного и ответственного использования инструментов с настраиваемыми лимитами.\n* ⚡ **Кэширование контента** : оптимизация производительности операций с файлами и URL-адресами с автоматическим обнаружением дубликатов.\n* 🔄 **Последовательное мышление** : обеспечивает многошаговые процессы рассуждения с отслеживанием прогресса.\n* 🔎 **Саморефлексия** : автоматизированные этапы рефлексии для улучшения качества рассуждений.\n* 📊 **Структурированные результаты** : автоматическое форматирование мыслительных процессов для лучшей читаемости.\n* 🔗 **Интеграция исследований** : бесшовное включение результатов веб-исследований в потоки рассуждений.\n\n## Преимущества структурированного мышления\n\nИспользование инструмента `think` обеспечивает специальное пространство для систематических рассуждений, поощряя:\n\n* Четкое определение проблемы\n* Сбор соответствующего контекста\n* Пошаговый анализ\n* Самоанализ рассуждений\n* Правильно сформулированные выводы\n\nНедавние исследования подчеркивают значительные улучшения при использовании структурированного мышления:\n\n* **Относительное улучшение на 54%** при выполнении сложных задач по принятию решений.\n* **Повышенная согласованность** в ходе нескольких испытаний.\n* **Улучшение производительности** в тестах по программной инженерии.\n\n## Подробные характеристики\n\nПомимо основного списка, MCP Think Tank предлагает сложные возможности для расширенного взаимодействия с ИИ.\n\n### Структурированное мышление (инструмент мышления)\n\nИнструмент `think` — это основной механизм для обеспечения продвинутого рассуждения ИИ. Он предоставляет специализированную структурированную среду, в которой ИИ может систематически разбивать проблемы, собирать контекст, анализировать варианты и выполнять самоанализ. Это способствует более глубокому анализу и более качественным результатам по сравнению с неструктурированными ответами. Он поддерживает последовательные шаги и легко интегрируется с инструментами исследования и памяти.\n\n#### Функция саморефлексии\n\nИнструмент «Мышление» включает в себя мощную возможность саморефлексии, которую можно включить с помощью параметра `selfReflect: true` :\n\n```javascript\nmcp_think-tool_think({\n structuredReasoning: \"...\",\n selfReflect: true,\n reflectPrompt: \"Optional custom reflection prompt\"\n})\n```\n\nКогда включена саморефлексия, ИИ получает подсказку для размышления над собственными рассуждениями. Это соответствует философии проектирования MCP, которая заключается в улучшении, а не замене возможностей ИИ.\n\nПараметр `reflectPrompt` позволяет вам настраивать подсказку, используемую для рефлексии, подгоняя ее под конкретные задачи или области рассуждений. Если не указано, используется подсказка по умолчанию, которая запрашивает идентификацию несоответствий, логических ошибок и предложения по улучшению.\n\n### Память графа знаний\n\nГраф знаний обеспечивает постоянную память в различных взаимодействиях и сеансах. Он позволяет ИИ строить растущее понимание проекта, его компонентов и связанных концепций.\n\n* **Наблюдения с меткой времени** : все записи памяти включают метаданные для отслеживания.\n* **Предотвращение дубликатов** : интеллектуальное сопоставление сущностей позволяет избежать избыточных записей.\n* **Автоматическое связывание** : создание отношений на основе эвристики связывает связанные концепции (настраивается).\n* **Расширенные запросы** : Фильтруйте память по времени, тегам, ключевым словам и т. д. с помощью мощного инструмента `memory_query` для исторического анализа и отслеживания эволюции концепций. Легко находите последние записи за последние 48 часов или за любой определенный период времени.\n* **Обслуживание памяти** : включены инструменты для обрезки и управления ростом памяти.\n* **Ключевые инструменты памяти** : такие инструменты, как `upsert_entities` , `add_observations` , `create_relations` , `search_nodes` , `memory_query` и `open_nodes` используются для взаимодействия с графом.\n\n### Инструменты управления задачами\n\nНабор инструментов позволяет ИИ управлять задачами проекта непосредственно в потоке разговора. Это интегрирует планирование и выполнение с графом знаний, позволяя ИИ понимать статус и приоритеты проекта.\n\n#### Инструменты для выполнения ключевых задач\n\n* `plan_tasks` : создание нескольких задач одновременно с приоритетами и зависимостями\n* `list_tasks` : Фильтрация задач по статусу и приоритету\n* `next_task` : Получить задачу с наивысшим приоритетом и отметить ее как выполняемую\n* `complete_task` : Отметить задачи как выполненные\n* `update_tasks` : Обновить существующие задачи новой информацией\n\n### Инструменты веб-исследования (Exa)\n\nИспользуя API Exa, MCP Think Tank предоставляет инструменты для получения внешней информации. Это позволяет ИИ получать доступ к актуальной информации из сети, чтобы информировать свои рассуждения и предоставлять ответы из источников.\n\n* `exa_search` : выполнение веб-поиска на основе запроса.\n* `exa_answer` : Получите краткий, обоснованный ответ на фактический вопрос.\n\n**Примечание:** использование этих инструментов требует настройки ключа Exa API. См. раздел [«Конфигурация»](#configuration) .\n\n### Оркестровка инструментов и меры безопасности\n\nMCP Think Tank включает в себя комплексные функции, обеспечивающие ответственное и эффективное использование инструментов.\n\n* **Ограничения использования** : настраиваемое максимальное количество вызовов инструментов на одно взаимодействие пользователя ( `TOOL_LIMIT` , по умолчанию: 25). Ограничение учитывает только последовательные вызовы инструментов в одном сообщении пользователя и автоматически сбрасывается, когда пользователь отправляет новое сообщение.\n* **Автоматическое отслеживание** : все вызовы инструментов регистрируются и отслеживаются.\n* **Плавное снижение** : при достижении пределов система пытается вернуть частичные результаты.\n* **Интеллектуальное кэширование** : идентичные вызовы инструментов и повторные выборки содержимого файлов/URL автоматически кэшируются, что сокращает время выполнения и использование ресурсов. Поведение и размер кэширования настраиваются ( `CACHE_TOOL_CALLS` , `CONTENT_CACHE` ).\n* **Настраиваемый доступ** : белый список инструментов может ограничивать доступные инструменты в определенных контекстах.\n* **Обработка ошибок** : надежная обработка ошибок обеспечивает четкую обратную связь в случае возникновения таких проблем, как достижение пределов или недопустимые вызовы инструментов.\n\n## 📦 Установка\n\n> ⚠️ **Важное примечание ПРОЧИТАЙТЕ ЭТО:** При обновлении до новой версии MCP Think Tank в Cursor или Claude вы можете создать несколько экземпляров сервера MCP Think Tank, что приведет к созданию дополнительных экземпляров Node.js, что снизит производительность вашей системы — это известная проблема с серверами MCP — завершите все процессы mcp-think-tank в вашей системе и убедитесь, что у вас запущен только один экземпляр node.js.\n\n> ⚠️ Tasks.jsonl находится в ~/.mcp-think-tank/. Файл отделен от файла kg, так как аналитический центр может запутаться в ранее созданных задачах в файле kg. Удалите содержимое файла tasks.jsonl, если файл станет слишком большим или если вы хотите начать новый проект и убедиться, что в файле нет задач. В будущей версии задачи могут быть объединены с файлом kg, чтобы гарантировать, что выполненные задачи и связи будут храниться в памяти, и не будет дублирующихся задач.\n\n### NPX (рекомендуется)\n\nСамый простой способ использовать MCP Think Tank — через NPX в Cursor, используя файл mcp.json, который запускает последнюю версию без глобальной установки.\n\n```bash\nnpx mcp-think-tank@2.0.7\n```\n\nДля последней версии (могут возникнуть проблемы с совместимостью):\n\n```bash\nnpx mcp-think-tank@latest\n```\n\n> У некоторых пользователей возникают проблемы с npx @latest в Cursor. Если это так, попробуйте указать версию mcp-think-tank@2.0.7 в файле .json или установить ее глобально:\n\n### Глобальная установка\n\nДля постоянного инструмента командной строки:\n\n```bash\nnpm install -g mcp-think-tank\nmcp-think-tank\n```\n\n## ⚙️ Конфигурация\n\nMCP Think Tank настраивается в основном через переменные среды или через конфигурацию вашего клиента MCP (например, `.cursor/mcp.json` Cursor).\n\n### Быстрый старт: основная настройка\n\n1. **Установите MCP Think Tank** (см. Установка выше).\n2. **Получите свой ключ API Exa** (требуется для инструментов веб-поиска):\n * Зарегистрируйтесь на [exa.ai](https://exa.ai/) и скопируйте свой ключ API.\n3. **ВАЖНО! СЕРВЕРЫ STDIO УСТАРЕЛИ** . Индустрия MCP переходит на транспорты на основе HTTP. БУДУЩИЕ ОБНОВЛЕНИЯ НЕ БУДУТ ПОДДЕРЖИВАТЬ СЕРВЕРЫ STDIO.\n4. **Настройте свой сервер MCP** (для Cursor добавьте в `.cursor/mcp.json` ):\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"type\": \"streamable-http\",\n \"env\": {\n \"MEMORY_PATH\": \"/absolute/path/to/your/project/memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n }\n }\n }\n}\n```\n\n### Основные переменные\n\n* `MEMORY_PATH` : **Обязательно** . Абсолютный путь к файлу хранилища памяти. **Важно:** всегда устанавливайте уникальный `MEMORY_PATH` для каждого проекта, чтобы избежать конфликтов графа знаний между проектами. Если не указано, по умолчанию используется `~/.mcp-think-tank/memory.jsonl` .\n* `EXA_API_KEY` : **Требуется для инструментов веб-поиска Exa** . Ваш ключ API от [exa.ai](https://exa.ai/) .\n\n### Расширенная конфигурация\n\n* `TOOL_LIMIT` : Максимальное количество вызовов инструментов, разрешенных для одного взаимодействия пользователя (по умолчанию: `25` ). Счетчик автоматически сбрасывается с каждым новым сообщением пользователя, гарантируя, что вы можете сделать до 25 последовательных вызовов инструментов в рамках одного взаимодействия.\n* `CACHE_TOOL_CALLS` : включить/отключить кэширование идентичных вызовов инструментов (по умолчанию: `true` ).\n* `TOOL_CACHE_SIZE` : Максимальное количество кэшированных вызовов инструментов (по умолчанию: `100` ).\n* `CACHE_CONTENT` : включить/отключить кэширование на основе содержимого для операций с файлами/URL (по умолчанию: `true` ).\n* `CONTENT_CACHE_SIZE` : Максимальное количество элементов в кэше контента (по умолчанию: `50` ).\n* `CONTENT_CACHE_TTL` : время жизни кэшированного контента в миллисекундах (по умолчанию: `300000` - 5 минут).\n* `MCP_DEBUG` : включить ведение журнала отладки (по умолчанию: `false` ).\n* `MCP_LISTEN_PORT` : Установить пользовательский порт для сервера MCP (по умолчанию: `3399` для серверов TCP, не имеет значения для `stdio` ).\n* `LOG_LEVEL` : Установить уровень ведения журнала ( `debug` , `info` , `warn` , `error` ) (по умолчанию: `info` ).\n* `AUTO_LINK` : включить автоматическое связывание сущностей в графе знаний (по умолчанию: `true` ).\n\n### Обслуживание памяти\n\n* `MIN_SIMILARITY_SCORE` : Пороговое значение для сопоставления сущностей при предотвращении дубликатов (по умолчанию: `0.85` ).\n* `MAX_OPERATION_TIME` : Максимальное время пакетных операций с памятью в миллисекундах (по умолчанию: `5000` ).\n\nПример конфигурации с расширенными настройками в `.cursor/mcp.json` :\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"env\": {\n \"MEMORY_PATH\": \"./project-memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n \"TOOL_LIMIT\": \"50\",\n \"CACHE_CONTENT\": \"true\",\n \"CONTENT_CACHE_SIZE\": \"100\",\n \"MCP_DEBUG\": \"false\",\n \"AUTO_LINK\": \"true\"\n }\n }\n }\n}\n```\n\n> 💡 **Совет по производительности:** для больших проектов увеличение `TOOL_LIMIT` и размеров кэша может повысить производительность за счет более высокого использования памяти. Отслеживайте свои шаблоны использования и вносите соответствующие изменения. Но в Cursor предел инструмента должен быть 25, чтобы избежать достижения предела и возобновления с последнего вызова инструмента — в настоящее время многие пользователи курсора сообщают о проблемах с возобновлением в версии: 0.49.6. Это не связано с MCP Think Tank.\n\n> 💡 **Примечание:** Если вы используете Cursor в режиме YOLO или кодировании Vibe, я предлагаю контекстную подготовку новых чатов и дать Cursor знать, что он должен использовать MCP Think Tank для создания сущностей, наблюдений и отношений. Это поможет вам получить максимальную отдачу от MCP Think Tank.\n\nПримером контекстного прайминга является сохранение файла `Prime.md` в папке `.cursor` вашего проекта со следующим содержимым:\n\n```markdown\n# Context Prime\n> Follow the instructions to understand the context of the project.\n\n## Run the following command\n\neza . --tree --git-ignore\n\n## Read the following files\n> Read the files below to get the context of the project. \n\n> list of files:\nREADME.md\n...\n\n## MCP Think Tank Tools\n> Test the MCP tools, first use 'show_memory_path' to remind the user of the current memory path file used, then use the 'memory_query' tool to find and read recent entities and observations for the last 48 hours so you are up to date.\n\n> Automatically utilize the MCP Think Tank to autonomously track project context, dynamically adding entities, observations, and relations to the knowledge graph while proactively querying for relevant information and historical insights. Use integrated planning and task management tools to enhance project efficiency. Keep track of the project and its context without the user having to ask for it.\n\n> Dont do anything else.\n```\n\nБолее подробную информацию о серверах MCP см. [в документации Cursor MCP](https://docs.cursor.com/context/model-context-protocol) .\n\n## Настройка правил проекта (для курсора/ИИ)\n\nЧтобы обеспечить эффективное использование Cursor и другими совместимыми агентами инструментов MCP Think Tank, вам необходимо предоставить ИИ руководство. Обычно это делается с помощью правила проекта. Создайте **одно правило проекта Auto Attach** следующим образом:\n\n### 1. Добавить новое правило в курсоре\n\n1. Открытый курсор.\n2. Перейдите в палитру команд ( `Cmd+Shift+P` или `Ctrl+Shift+P` ).\n3. Выберите **«Новое правило курсора»** .\n4. Назовите правило (например, `mcp-think-tank.mdc` ).\n5. В редакторе правил задайте метаданные и вставьте содержимое правила из примера ниже.\n\n### 2. Пример файла правил ( `.cursor/rules/mcp-think-tank.mdc` )\n\nЭтот файл Markdown служит контекстом *для ИИ* , помогая ему определить, когда и как использовать доступные инструменты.\n\n```mdc\nrule type: auto attach\nuse globs: **/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.md, **/*.py, **/*.json\n```\n\n\\----- Начало правила -----\n\n```markdown\nRegularly utilize MCP Think Tank tools to maintain an updated knowledge graph and maximize its potential. Simply call the tools in your prompt. \n\n## Quick Decision Tree\n\n1. 🤔 **Complex problem to analyze?**\n → Use `think` to structure reasoning and reflect\n\n2. 🔍 **Need past context or information?**\n → Use `memory_query` (time-based) or `search_nodes` (keyword-based)\n\n3. 📊 **Planning implementation steps?**\n → Use `plan_tasks` to create and track work\n\n4. 🌐 **Need current external information?**\n → Use `exa_search` (general search) or `exa_answer` (factual questions)\n\n## Critical Memory Management (Automatic Use Required)\n\n| When to automatically use memory | Tool to use |\n|------------------|------------|\n| At session start | `memory_query` with recent time filter (last 24h) |\n| After completing significant analysis | `upsert_entities` to store conclusions |\n| When context seems missing | `memory_query` with relevant keyword |\n| Every ~30 minutes in long sessions | `upsert_entities` to create checkpoint |\n| When switching between major topics | `think` + `upsert_entities` to summarize progress |\n| Before session end | `upsert_entities` to store session summary |\n\n## Core Workflows\n\n### Workflow 1: Problem Analysis → Solution\n1. `memory_query` → Check for relevant past work\n2. `think` → Structure reasoning about the problem \n3. `plan_tasks` → Break down implementation steps\n4. `upsert_entities` → Store conclusions in memory\n\n### Workflow 2: Research → Knowledge Capture\n1. `memory_query` → Check if already researched\n2. `exa_search` → Find current information\n3. `think` → Analyze findings\n4. `upsert_entities` → Document key concepts\n\n### Workflow 3: Context Recovery (Session Resume)\n1. `memory_query` → Retrieve recent work (past 24-48h)\n2. `open_nodes` → Get details on relevant entities\n3. `think` → Synthesize context and plan next steps\n4. Continue where left off\n\n### Workflow 4: Task Management\n1. `list_tasks` → Review current work status\n2. `next_task` → Identify priority task\n3. `complete_task` → Mark finished work\n4. `upsert_entities` → Document completion\n\n## Trigger Patterns (Automatic Tool Use)\n\n| When user... | Automatically use... |\n|--------------|----------------------|\n| Asks complex question requiring analysis | `think` |\n| Mentions \"remember\" or refers to past work | `memory_query` with time filter → `open_nodes` |\n| Uses \"research\" or \"find latest\" | `memory_query` (check if already known) → `exa_search` |\n| Asks factual questions needing citations | `exa_answer` |\n| Mentions planning or implementation | `plan_tasks` |\n| Refers to continuing previous work | `memory_query` → `list_tasks` → `next_task` |\n| Seems to have lost context from earlier | `memory_query` with recent time filter |\n| Makes significant conceptual progress | `upsert_entities` without being asked |\n| Connects related concepts | `create_relations` |\n| Completes major section of work | `think` + `upsert_entities` to summarize |\n\n### When To Use Each Memory Tool\n\n- `memory_query`: For time-based searches and recent context recovery\n- `search_nodes`: For finding specific concepts by keyword\n- `open_nodes`: For retrieving full details of known entities\n- `upsert_entities`: For creating new knowledge or updating existing entities\n- `add_observations`: For adding facts to existing entities\n- `create_relations`: For connecting related concepts\n\n## Other Tools Reference\n\n### Thinking\n- `think`: Structured reasoning with optional reflection\n\n### Tasks\n- `plan_tasks`: Create task list\n- `list_tasks`: View current tasks\n- `next_task`: Get priority task\n- `complete_task`: Mark task done\n\n### Research\n- `exa_search`: Web search\n- `exa_answer`: Get cited answers\n\n## AI Behavior Requirements\n\n1. ALWAYS check memory at session start with `memory_query`\n2. AUTOMATICALLY store important conclusions with `upsert_entities`\n3. CREATE periodic memory checkpoints during long sessions\n4. PROACTIVELY check memory when context seems missing\n5. CHAIN tools together following the workflows\n6. PRIORITIZE memory tools before web research\n7. SUMMARIZE progress before ending major work segments\n```\n\n\\----- Конец правила -----\n\n## ⚡ Оптимизация производительности\n\nMCP Think Tank использует встроенные оптимизации для обеспечения эффективной работы:\n\n### Кэширование контента\n\n* Автоматическое кэширование содержимого файлов и URL на основе криптографического хеширования.\n* Предотвращает избыточное чтение файлов и сетевые запросы.\n* Значительно ускоряет повторные операции с одним и тем же контентом.\n* Размер кэша и TTL настраиваются с помощью переменных среды ( `CONTENT_CACHE_SIZE` , `CONTENT_CACHE_TTL` ).\n\n### Оптимизация вызова инструмента\n\n* Идентичные вызовы инструментов в рамках сеанса автоматически обнаруживаются и обслуживаются из кэша.\n* Предотвращает подсчет повторных вызовов в рамках лимита взаимодействия.\n* Улучшает реакцию на повторяющиеся запросы инструментов.\n* Размер кэша можно настраивать ( `TOOL_CACHE_SIZE` ).\n\n## Лучшие практики\n\nДля оптимального использования MCP Think Tank с Cursor/Claude в крупных проектах:\n\n* Используйте инструмент `think` для всех нетривиальных процессов рассуждения и принятия решений.\n* Всегда сохраняйте важные мысли, выводы и архитектурные решения в графе знаний, используя инструменты памяти.\n* Интегрируйте веб-исследования и управление задачами в свой рабочий процесс, чтобы ИИ был информирован и сосредоточен.\n* Регулярно просматривайте и обновляйте схему знаний вашего проекта, чтобы обеспечить ее точность и актуальность.\n* Ссылайтесь на существующие знания и прошлые решения, чтобы поддерживать последовательность кода и дизайна.\n* Помните об ограничениях на вызов инструментов, особенно в сложных автоматизированных рабочих процессах. Контролируйте использование, если необходимо.\n* Настройте переменные конфигурации ( `TOOL_LIMIT` , параметры кэша) в зависимости от потребностей и сложности вашего проекта для повышения производительности.\n\n## 🤝 Вклад\n\nВклады приветствуются! Пожалуйста, не стесняйтесь отправлять запрос на включение.\n\n1. Создайте форк репозитория.\n2. Создайте ветку функций ( `git checkout -b feature/amazing-feature` ).\n3. Зафиксируйте свои изменения ( `git commit -m 'Add some amazing feature'` ).\n4. Отправьте изменения в ветку ( `git push origin feature/amazing-feature` ).\n5. Откройте запрос на извлечение.\n\n## 📄 Лицензия\n\nДанный проект лицензирован по лицензии MIT — подробности см. в файле [LICENSE](LICENSE) .\n\n### 📚 Ссылки для справки\n\n* [Документация по правилам курсора](https://docs.cursor.com/context/rules)\n* [Протокол контекста модели MCP](https://docs.cursor.com/context/model-context-protocol)\n* [API Exa](https://exa.ai/)\n* [Исследования структурированного мышления компании Anthropic](https://www.anthropic.com/research)\n* [Модель контекстного протокола](https://github.com/modelcontextprotocol/typescript-sdk)\n* [FastMCP](https://github.com/jlowin/fastmcp)\n\n***","Russian","ru-RU",{"_29":76,"_33":80,"_38":81},{"_31":77,"_33":78,"_35":79},"通过结构化思维、持久知识图谱记忆和智能工具编排，为AI助手提供增强的推理能力，以解决复杂的问题。","MCP智库","# MCP智库\n\n[![npm 版本](https://img.shields.io/npm/v/mcp-think-tank.svg?style=flat-square)](https://www.npmjs.com/package/mcp-think-tank) [![许可证：MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![TypeScript](https://img.shields.io/badge/TypeScript-5.8-blue.svg)](https://www.typescriptlang.org/) [![兼容光标](https://img.shields.io/badge/Cursor-Compatible-4B8DF8?logo=cursor\\&logoColor=white\\&style=flat-square)](https://www.cursor.so/) [![克劳德兼容性](https://img.shields.io/badge/Claude-Compatible-9370DB.svg)](https://www.anthropic.com/) [![模型上下文协议](https://img.shields.io/badge/MCP-Tank-orange.svg)](https://github.com/modelcontextprotocol) [![Node.js](https://img.shields.io/badge/Node.js-18+-green.svg)](https://nodejs.org/) [![铁匠徽章](https://smithery.ai/badge/@flight505/mcp-think-tank)](https://smithery.ai/server/@flight505/mcp-think-tank)\n\n## 概述\n\nMCP Think Tank 是一个强大的模型上下文协议 (MCP) 服务器，旨在增强**Cursor**和**Claude @Web**等 AI 助手的功能。它为增强推理、持久记忆和负责任的工具使用提供了一个结构化的环境。\n\n关键功能包括先进的**顺序思维和链式推理**、具有版本控制功能的强大**知识图谱存储**系统，以及**具有调用限制保障措施的智能工具编排**。该平台使人工智能能够通过结构化分析解决复杂问题，维护跨会话的知识，并利用网络搜索等外部资源，同时遵守可配置的使用限制。\n\n## 🎯 哲学\n\nMCP智库建立在三个核心原则之上：\n\n1. **优雅简约**：简约、精心设计的工具，可以补充人工智能功能，而不是试图复制它们。\n2. **增强反思**：温和的指导可以促进更好的推理和自我反思，而不受严格的约束。\n3. **持久上下文**：简单但有效的知识图谱提供了跨对话的记忆。\n\n## 主要特点\n\n* 💭**思考工具**：用于结构化推理和自我反思的专用空间。\n* 🧩**知识图谱**：简单有效的持久记忆系统。\n* 📝**任务管理工具**：计划、跟踪和更新任务，与知识图谱集成。\n* 🌐 **Web 研究工具 (Exa)** ：使用 Exa API 搜索网络并获取来源答案。\n* 🔍**记忆工具**：用于存储和检索知识图谱中信息的易于使用的工具。\n* 🤝**客户端支持**：与 Cursor、Claude @Web 和其他 MCP 客户端无缝集成。\n* 🛡️**工具编排和调用限制**：内置安全措施，通过可配置的限制实现高效、负责任的工具使用。\n* ⚡**内容缓存**：通过自动重复检测对文件和 URL 操作进行性能优化。\n* 🔄**顺序思维**：实现具有进度跟踪的多步骤推理过程。\n* 🔎**自我反思**：自动反思以提高推理质量。\n* 📊**结构化输出**：自动格式化思维过程以提高可读性。\n* 🔗**研究整合**：将网络研究结果无缝融入推理流程。\n\n## 结构化思维的好处\n\n利用`think`工具为系统推理提供专用空间，鼓励：\n\n* 明确问题定义\n* 相关背景收集\n* 逐步分析\n* 自我反思推理\n* 完善的结论\n\n最近的研究强调了使用结构化思维带来的显著改善：\n\n* 复杂决策任务**相对提高54%** 。\n* **增强多次试验的一致性**。\n* **提高了软件工程基准测试的性能**。\n\n## 详细功能\n\n除了核心列表之外，MCP Think Tank 还提供了用于高级 AI 交互的复杂功能。\n\n### 结构化思维（思考工具）\n\n`think`工具是实现高级人工智能推理的核心机制。它提供了一个专用的结构化环境，让人工智能能够系统地分解问题、收集上下文、分析选项并进行自我反思。与非结构化的响应相比，这能够促进更深入的分析和更高质量的输出。它支持顺序步骤，并与研究和记忆工具无缝集成。\n\n#### 自我反思功能\n\n思考工具包含强大的自我反思功能，可以通过`selfReflect: true`参数启用：\n\n```javascript\nmcp_think-tool_think({\n structuredReasoning: \"...\",\n selfReflect: true,\n reflectPrompt: \"Optional custom reflection prompt\"\n})\n```\n\n启用自我反思功能后，AI 会收到反思自身推理的提示。这符合 MCP 的设计理念，即增强而非取代 AI 能力。\n\n您可以使用`reflectPrompt`参数自定义用于反射的提示，使其适用于特定的推理任务或领域。如果未指定，则使用默认提示，要求识别不一致之处、逻辑错误并提供改进建议。\n\n### 知识图谱记忆\n\n知识图谱为不同的交互和会话提供持久记忆。它使人工智能能够逐渐加深对项目、其组成部分和相关概念的理解。\n\n* **带时间戳的观察**：所有内存条目都包含用于跟踪的元数据。\n* **重复预防**：智能实体匹配避免冗余条目。\n* **自动链接**：基于启发式的关系创建连接相关概念（可配置）。\n* **高级查询**：使用强大的`memory_query`工具，按时间、标签、关键词等筛选记忆，进行历史分析并追踪概念演变。轻松查找过去48小时或任何特定时间段内的最新条目。\n* **内存维护**：包括修剪和管理内存增长的工具。\n* **关键内存工具**： `upsert_entities` 、 `add_observations` 、 `create_relations` 、 `search_nodes` 、 `memory_query`和`open_nodes`等工具用于与图表交互。\n\n### 任务管理工具\n\n一套工具允许人工智能直接在对话流程中管理项目任务。这将规划和执行与知识图谱整合在一起，使人工智能能够了解项目状态和优先级。\n\n#### 关键任务工具\n\n* `plan_tasks` ：同时创建具有优先级和依赖关系的多个任务\n* `list_tasks` ：按状态和优先级过滤任务\n* `next_task` ：获取最高优先级的任务并将其标记为正在进行中\n* `complete_task` ：将任务标记为已完成\n* `update_tasks` ：使用新信息更新现有任务\n\n### 网络研究工具（Exa）\n\n利用 Exa API，MCP Think Tank 提供了获取外部信息的工具。这使得人工智能能够从网络上访问最新信息，为其推理提供依据并提供可靠的答案。\n\n* `exa_search` ：根据查询执行网络搜索。\n* `exa_answer` ：获得对事实问题的简洁、有来源的答案。\n\n\\*\\*注意：\\*\\*使用这些工具需要配置您的 Exa API 密钥。请参阅[配置](#configuration)部分。\n\n### 工具编排与保障\n\nMCP Think Tank 包含全面的功能，以确保负责任且高效地使用工具。\n\n* **使用限制**：可配置每次用户交互的最大工具调用次数（ `TOOL_LIMIT` ，默认值：25）。此限制仅计算单条用户消息中的连续工具调用次数，并在用户发送新消息时自动重置。\n* **自动跟踪**：所有工具调用均被记录和监控。\n* **优雅降级**：当达到限制时，系统会尝试返回部分结果。\n* **智能缓存**：相同的工具调用和重复的文件/URL 内容获取将自动缓存，从而减少执行时间和资源占用。缓存行为和大小可配置（ `CACHE_TOOL_CALLS` 、 `CONTENT_CACHE` ）。\n* **可配置访问**：工具白名单可以限制特定环境中可用的工具。\n* **错误处理**：强大的错误处理为达到限制或无效工具调用等问题提供清晰的反馈。\n\n## 📦安装\n\n> ⚠️\\*\\*重要提示阅读此内容：\\*\\*在 Cursor 或 Claude 中更新到新版本的 MCP Think Tank 时，您可能会创建 MCP Think Tank 服务器的多个实例，从而导致创建额外的 Node.js 实例，从而降低系统性能 - 这是 MCP 服务器的已知问题 - 请终止系统中的所有 mcp-think-tank 进程，并检查是否只有一个 node.js 实例正在运行。\n\n> ⚠️ task.jsonl 位于 ~/.mcp-think-tank/ 目录。该文件与 kg 文件分离，因为智库可能会与 kg 文件中先前创建的任务混淆。如果文件过大，或者您想启动新项目并确保文件中没有任务，请删除 task.jsonl 文件的内容。在未来的版本中，任务可能会与 kg 文件合并，以确保已完成的任务和关系存储在内存中，并且不会出现重复的任务。\n\n### NPX（推荐）\n\n使用 MCP Think Tank 最简单的方法是通过 Cursor 中的 NPX 使用 mcp.json 文件，该文件无需全局安装即可运行最新版本，\n\n```bash\nnpx mcp-think-tank@2.0.7\n```\n\n对于最新版本（可能存在兼容性问题）：\n\n```bash\nnpx mcp-think-tank@latest\n```\n\n> 一些用户在 Cursor 中使用 npx @latest 时遇到问题，如果是这样，请尝试在 .json 文件中指定版本 mcp-think-tank@2.0.7，或者全局安装它：\n\n### 全局安装\n\n对于持久命令行工具：\n\n```bash\nnpm install -g mcp-think-tank\nmcp-think-tank\n```\n\n## ⚙️ 配置\n\nMCP Think Tank 主要通过环境变量或通过 MCP 客户端的配置（如 Cursor 的`.cursor/mcp.json` ）进行配置。\n\n### 快速入门：基本设置\n\n1. **安装 MCP Think Tank** （参见上面的安装）。\n2. **获取您的 Exa API 密钥**（网络搜索工具所需）：\n * 在[exa.ai](https://exa.ai/)注册并复制您的 API 密钥。\n3. **重要的 STDIO 服务器已被弃用**- MCP 行业正在转向基于 HTTP 的传输，- 未来的更新将不支持 STDIO 服务器。\n4. **配置您的 MCP 服务器**（对于 Cursor，添加到`.cursor/mcp.json` ）：\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"type\": \"streamable-http\",\n \"env\": {\n \"MEMORY_PATH\": \"/absolute/path/to/your/project/memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n }\n }\n }\n}\n```\n\n### 基本变量\n\n* `MEMORY_PATH` ：**必需**。内存存储文件的绝对路径。\\*\\*重要提示：\\*\\*务必为每个项目设置唯一的`MEMORY_PATH` ，以避免项目间的知识图谱冲突。如果省略，则默认为`~/.mcp-think-tank/memory.jsonl` 。\n* `EXA_API_KEY` ： **Exa 网络搜索工具所需**。您从[exa.ai](https://exa.ai/)获取的 API 密钥。\n\n### 高级配置\n\n* `TOOL_LIMIT` ：每次用户交互允许的最大工具调用次数（默认值： `25` ）。计数器会随着每条新的用户消息自动重置，确保您在单次交互中最多可以连续调用 25 次工具。\n* `CACHE_TOOL_CALLS` ：启用/禁用相同工具调用的缓存（默认值： `true` ）。\n* `TOOL_CACHE_SIZE` ：缓存工具调用的最大数量（默认值： `100` ）。\n* `CACHE_CONTENT` ：启用/禁用文件/URL 操作的基于内容的缓存（默认值： `true` ）。\n* `CONTENT_CACHE_SIZE` ：内容缓存中的最大项目数（默认值： `50` ）。\n* `CONTENT_CACHE_TTL` ：缓存内容的生存时间（毫秒）（默认值： `300000` - 5 分钟）。\n* `MCP_DEBUG` ：启用调试日志记录（默认值： `false` ）。\n* `MCP_LISTEN_PORT` ：为 MCP 服务器设置自定义端口（默认值：TCP 服务器为`3399` ，与`stdio`无关）。\n* `LOG_LEVEL` ：设置日志级别（ `debug` 、 `info` 、 `warn` 、 `error` ）（默认值： `info` ）。\n* `AUTO_LINK` ：在知识图谱中启用自动实体链接（默认值： `true` ）。\n\n### 记忆维护\n\n* `MIN_SIMILARITY_SCORE` ：防止重复时的实体匹配阈值（默认值： `0.85` ）。\n* `MAX_OPERATION_TIME` ：批量内存操作的最长时间（以毫秒为单位）（默认值： `5000` ）。\n\n`.cursor/mcp.json`中的高级设置示例配置：\n\n```json\n{\n \"mcpServers\": {\n \"think-tank\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-think-tank@2.0.7\"],\n \"env\": {\n \"MEMORY_PATH\": \"./project-memory.jsonl\",\n \"EXA_API_KEY\": \"your-exa-api-key-here\",\n \"TOOL_LIMIT\": \"50\",\n \"CACHE_CONTENT\": \"true\",\n \"CONTENT_CACHE_SIZE\": \"100\",\n \"MCP_DEBUG\": \"false\",\n \"AUTO_LINK\": \"true\"\n }\n }\n }\n}\n```\n\n> 💡\\*\\*性能提示：\\*\\*对于大型项目，增加`TOOL_LIMIT`和缓存大小可以提升性能，但会增加内存占用。请监控您的使用模式并进行相应调整。但在 Cursor 中，工具限制应为 25，以避免达到限制并从上次工具调用中恢复执行 - 目前许多 Cursor 用户报告在 0.49.6 版本中存在恢复问题。这与 MCP Think Tank 无关。\n\n> 💡\\*\\*注意：\\*\\*如果您在 YOLO 模式或 Vibe 编码中使用 Cursor，我建议您启动新的聊天上下文，并告知 Cursor 应该使用 MCP Think Tank 来创建实体、观察值和关系。这将帮助您充分利用 MCP Think Tank。\n\n上下文启动的一个例子是，在项目的`.cursor`文件夹中保留一个`Prime.md`文件，其中包含以下内容：\n\n```markdown\n# Context Prime\n> Follow the instructions to understand the context of the project.\n\n## Run the following command\n\neza . --tree --git-ignore\n\n## Read the following files\n> Read the files below to get the context of the project. \n\n> list of files:\nREADME.md\n...\n\n## MCP Think Tank Tools\n> Test the MCP tools, first use 'show_memory_path' to remind the user of the current memory path file used, then use the 'memory_query' tool to find and read recent entities and observations for the last 48 hours so you are up to date.\n\n> Automatically utilize the MCP Think Tank to autonomously track project context, dynamically adding entities, observations, and relations to the knowledge graph while proactively querying for relevant information and historical insights. Use integrated planning and task management tools to enhance project efficiency. Keep track of the project and its context without the user having to ask for it.\n\n> Dont do anything else.\n```\n\n有关 MCP 服务器的更多详细信息，请参阅[Cursor MCP 文档](https://docs.cursor.com/context/model-context-protocol)。\n\n## 项目规则设置（用于光标/AI）\n\n为了确保 Cursor 和其他兼容代理能够有效利用 MCP Think Tank 的工具，您需要为 AI 提供指导。这通常通过项目规则来实现。请按如下方式创建**一条“自动附加”项目规则**：\n\n### 1. 在 Cursor 中添加新规则\n\n1. 打开光标。\n2. 转到命令面板（ `Cmd+Shift+P`或`Ctrl+Shift+P` ）。\n3. 选择\\*\\*“新建光标规则”\\*\\* 。\n4. 命名规则（例如， `mcp-think-tank.mdc` ）。\n5. 在规则编辑器中，设置元数据并粘贴以下示例中的规则内容。\n\n### 2. 示例规则文件（ `.cursor/rules/mcp-think-tank.mdc` ）\n\n这个 Markdown 文件作为*AI 的*上下文，指导它何时以及如何使用可用的工具。\n\n```mdc\nrule type: auto attach\nuse globs: **/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.md, **/*.py, **/*.json\n```\n\n\\----- 规则开始 -----\n\n```markdown\nRegularly utilize MCP Think Tank tools to maintain an updated knowledge graph and maximize its potential. Simply call the tools in your prompt. \n\n## Quick Decision Tree\n\n1. 🤔 **Complex problem to analyze?**\n → Use `think` to structure reasoning and reflect\n\n2. 🔍 **Need past context or information?**\n → Use `memory_query` (time-based) or `search_nodes` (keyword-based)\n\n3. 📊 **Planning implementation steps?**\n → Use `plan_tasks` to create and track work\n\n4. 🌐 **Need current external information?**\n → Use `exa_search` (general search) or `exa_answer` (factual questions)\n\n## Critical Memory Management (Automatic Use Required)\n\n| When to automatically use memory | Tool to use |\n|------------------|------------|\n| At session start | `memory_query` with recent time filter (last 24h) |\n| After completing significant analysis | `upsert_entities` to store conclusions |\n| When context seems missing | `memory_query` with relevant keyword |\n| Every ~30 minutes in long sessions | `upsert_entities` to create checkpoint |\n| When switching between major topics | `think` + `upsert_entities` to summarize progress |\n| Before session end | `upsert_entities` to store session summary |\n\n## Core Workflows\n\n### Workflow 1: Problem Analysis → Solution\n1. `memory_query` → Check for relevant past work\n2. `think` → Structure reasoning about the problem \n3. `plan_tasks` → Break down implementation steps\n4. `upsert_entities` → Store conclusions in memory\n\n### Workflow 2: Research → Knowledge Capture\n1. `memory_query` → Check if already researched\n2. `exa_search` → Find current information\n3. `think` → Analyze findings\n4. `upsert_entities` → Document key concepts\n\n### Workflow 3: Context Recovery (Session Resume)\n1. `memory_query` → Retrieve recent work (past 24-48h)\n2. `open_nodes` → Get details on relevant entities\n3. `think` → Synthesize context and plan next steps\n4. Continue where left off\n\n### Workflow 4: Task Management\n1. `list_tasks` → Review current work status\n2. `next_task` → Identify priority task\n3. `complete_task` → Mark finished work\n4. `upsert_entities` → Document completion\n\n## Trigger Patterns (Automatic Tool Use)\n\n| When user... | Automatically use... |\n|--------------|----------------------|\n| Asks complex question requiring analysis | `think` |\n| Mentions \"remember\" or refers to past work | `memory_query` with time filter → `open_nodes` |\n| Uses \"research\" or \"find latest\" | `memory_query` (check if already known) → `exa_search` |\n| Asks factual questions needing citations | `exa_answer` |\n| Mentions planning or implementation | `plan_tasks` |\n| Refers to continuing previous work | `memory_query` → `list_tasks` → `next_task` |\n| Seems to have lost context from earlier | `memory_query` with recent time filter |\n| Makes significant conceptual progress | `upsert_entities` without being asked |\n| Connects related concepts | `create_relations` |\n| Completes major section of work | `think` + `upsert_entities` to summarize |\n\n### When To Use Each Memory Tool\n\n- `memory_query`: For time-based searches and recent context recovery\n- `search_nodes`: For finding specific concepts by keyword\n- `open_nodes`: For retrieving full details of known entities\n- `upsert_entities`: For creating new knowledge or updating existing entities\n- `add_observations`: For adding facts to existing entities\n- `create_relations`: For connecting related concepts\n\n## Other Tools Reference\n\n### Thinking\n- `think`: Structured reasoning with optional reflection\n\n### Tasks\n- `plan_tasks`: Create task list\n- `list_tasks`: View current tasks\n- `next_task`: Get priority task\n- `complete_task`: Mark task done\n\n### Research\n- `exa_search`: Web search\n- `exa_answer`: Get cited answers\n\n## AI Behavior Requirements\n\n1. ALWAYS check memory at session start with `memory_query`\n2. AUTOMATICALLY store important conclusions with `upsert_entities`\n3. CREATE periodic memory checkpoints during long sessions\n4. PROACTIVELY check memory when context seems missing\n5. CHAIN tools together following the workflows\n6. PRIORITIZE memory tools before web research\n7. SUMMARIZE progress before ending major work segments\n```\n\n\\----- 规则结束 -----\n\n## ⚡ 性能优化\n\nMCP Think Tank 采用内置优化功能，确保高效运行：\n\n### 内容缓存\n\n* 基于加密散列自动缓存文件和 URL 内容。\n* 防止冗余文件读取和网络请求。\n* 大大加快了对相同内容的重复操作。\n* 缓存大小和TTL可以通过环境变量（ `CONTENT_CACHE_SIZE` ， `CONTENT_CACHE_TTL` ）配置。\n\n### 工具调用优化\n\n* 会话中的相同工具调用会被自动检测并从缓存中提供服务。\n* 防止根据交互限制计算重复调用。\n* 提高对重复工具请求的响应能力。\n* 缓存大小是可配置的（ `TOOL_CACHE_SIZE` ）。\n\n## 最佳实践\n\n为了在大型项目中最佳地使用 MCP Think Tank 和 Cursor/Claude：\n\n* 利用`think`工具进行所有非平凡的推理和决策过程。\n* 始终使用记忆工具将重要的想法、结论和架构决策保留到知识图谱中。\n* 将网络研究和任务管理集成到您的工作流程中，以使 AI 保持知情和专注。\n* 定期审查和更新项目的知识图谱，以确保其准确性和相关性。\n* 参考现有知识和过去的决策来保持代码和设计的一致性。\n* 注意工具调用限制，尤其是在复杂的自动化工作流程中。如有必要，请监控使用情况。\n* 根据项目需求和复杂性调整配置变量（ `TOOL_LIMIT` 、缓存设置），以获得更好的性能。\n\n## 🤝 贡献\n\n欢迎贡献代码！欢迎提交 Pull 请求。\n\n1. 分叉存储库。\n2. 创建你的功能分支（ `git checkout -b feature/amazing-feature` ）。\n3. 提交您的更改（ `git commit -m 'Add some amazing feature'` ）。\n4. 推送到分支（ `git push origin feature/amazing-feature` ）。\n5. 打开拉取请求。\n\n## 📄 许可证\n\n该项目根据 MIT 许可证获得许可 - 有关详细信息，请参阅[LICENSE](LICENSE)文件。\n\n### 📚 参考链接\n\n* [游标规则文档](https://docs.cursor.com/context/rules)\n* [MCP 模型上下文协议](https://docs.cursor.com/context/model-context-protocol)\n* [Exa API](https://exa.ai/)\n* [人思结构化思维研究](https://www.anthropic.com/research)\n* [模型上下文协议](https://github.com/modelcontextprotocol/typescript-sdk)\n* [FastMCP](https://github.com/jlowin/fastmcp)\n\n***","Chinese","zh-CN","githubUser","locale",{"_85":86,"_87":88,"_13":153,"_161":42,"_162":42,"_163":164,"_167":168,"_190":-5,"_16":191,"_192":193,"_194":195,"_33":43,"_219":180,"_220":221,"_224":-5,"_225":226,"_227":228,"_235":236,"_241":-5,"_202":176,"_242":243,"_247":248,"_181":640,"_641":642},"addedAt","2025-05-08T05:01:54.342342376Z","argumentsJsonSchema",{"_89":90,"_150":151,"_96":152},"properties",{"_91":92,"_98":99,"_102":103,"_106":107,"_110":111,"_113":114,"_117":118,"_120":121,"_123":124,"_127":128,"_131":132,"_134":135,"_138":139,"_142":143,"_146":147},"AUTO_LINK",{"_93":94,"_31":95,"_96":97},"default","true","Enable automatic entity linking in knowledge graph","type","string","LOG_LEVEL",{"_93":100,"_31":101,"_96":97},"info","Set logging level (debug, info, warn, error)","MCP_DEBUG",{"_93":104,"_31":105,"_96":97},"false","Enable debug logging","TOOL_LIMIT",{"_93":108,"_31":109,"_96":97},"25","Maximum number of tool calls per session","EXA_API_KEY",{"_31":112,"_96":97},"API key for Exa web search (required for exa_search and exa_answer tools)","MEMORY_PATH",{"_93":115,"_31":116,"_96":97},"~/.mcp-think-tank/memory.jsonl","Path to the memory storage file","MCP_LOG_FILE",{"_93":94,"_31":119,"_96":97},"Enable/disable file logging","CACHE_CONTENT",{"_93":94,"_31":122,"_96":97},"Enable/disable content-based caching for file/URL operations","MCP_LISTEN_PORT",{"_93":125,"_31":126,"_96":97},"3399","Set custom port for MCP server","TOOL_CACHE_SIZE",{"_93":129,"_31":130,"_96":97},"100","Maximum number of cached tool calls","CACHE_TOOL_CALLS",{"_93":94,"_31":133,"_96":97},"Enable/disable duplicate tool call caching","CONTENT_CACHE_TTL",{"_93":136,"_31":137,"_96":97},"300000","Time-to-live for cached content in milliseconds","CONTENT_CACHE_SIZE",{"_93":140,"_31":141,"_96":97},"50","Maximum number of items in content cache","MAX_OPERATION_TIME",{"_93":144,"_31":145,"_96":97},"5000","Maximum time for batch operations in milliseconds","MIN_SIMILARITY_SCORE",{"_93":148,"_31":149,"_96":97},"0.85","Threshold for entity matching","required",[110],"object",[154,155,156,157,158,159,160],"category:knowledge-and-memory","category:autonomous-agents","category:agent-orchestration","language:typescript","security_grade:a","quality_grade:a","license_grade:a","descriptionMarkdown","descriptionPlainText","githubProject",{"_35":44,"_165":166},"url","https://github.com/flight505/mcp-think-tank","githubRepository",{"_169":170,"_171":172,"_173":-5,"_174":175,"_33":176,"_177":178,"_183":184,"_185":186,"_188":189},"defaultBranch","main","fullName","flight505/mcp-think-tank","githubTreeId","language","TypeScript","mcp-think-tank","owner",{"_179":180,"_181":182},"login","flight505","uid","gIo3bMycWC","recentStargazers",1,"spdxLicense",{"_33":187},"MIT License","stargazers",5,"glamaJson",6447,"inspectable",true,"integrations",[196,206,213],{"_197":198,"_31":204,"_181":205},"brand",{"_199":200,"_33":201,"_202":203},"logoSvgPath","M11.998,24c-0.321,0-0.641-0.084-0.922-0.247l-2.936-1.737c-0.438-0.245-0.224-0.332-0.08-0.383 c0.585-0.203,0.703-0.25,1.328-0.604c0.065-0.037,0.151-0.023,0.218,0.017l2.256,1.339c0.082,0.045,0.197,0.045,0.272,0l8.795-5.076 c0.082-0.047,0.134-0.141,0.134-0.238V6.921c0-0.099-0.053-0.192-0.137-0.242l-8.791-5.072c-0.081-0.047-0.189-0.047-0.271,0 L3.075,6.68C2.99,6.729,2.936,6.825,2.936,6.921v10.15c0,0.097,0.054,0.189,0.139,0.235l2.409,1.392 c1.307,0.654,2.108-0.116,2.108-0.89V7.787c0-0.142,0.114-0.253,0.256-0.253h1.115c0.139,0,0.255,0.112,0.255,0.253v10.021 c0,1.745-0.95,2.745-2.604,2.745c-0.508,0-0.909,0-2.026-0.551L2.28,18.675c-0.57-0.329-0.922-0.945-0.922-1.604V6.921 c0-0.659,0.353-1.275,0.922-1.603l8.795-5.082c0.557-0.315,1.296-0.315,1.848,0l8.794,5.082c0.57,0.329,0.924,0.944,0.924,1.603 v10.15c0,0.659-0.354,1.273-0.924,1.604l-8.794,5.078C12.643,23.916,12.324,24,11.998,24z M19.099,13.993 c0-1.9-1.284-2.406-3.987-2.763c-2.731-0.361-3.009-0.548-3.009-1.187c0-0.528,0.235-1.233,2.258-1.233 c1.807,0,2.473,0.389,2.747,1.607c0.024,0.115,0.129,0.199,0.247,0.199h1.141c0.071,0,0.138-0.031,0.186-0.081 c0.048-0.054,0.074-0.123,0.067-0.196c-0.177-2.098-1.571-3.076-4.388-3.076c-2.508,0-4.004,1.058-4.004,2.833 c0,1.925,1.488,2.457,3.895,2.695c2.88,0.282,3.103,0.703,3.103,1.269c0,0.983-0.789,1.402-2.642,1.402 c-2.327,0-2.839-0.584-3.011-1.742c-0.02-0.124-0.126-0.215-0.253-0.215h-1.137c-0.141,0-0.254,0.112-0.254,0.253 c0,1.482,0.806,3.248,4.655,3.248C17.501,17.007,19.099,15.91,19.099,13.993z","Node.js","slug","node-dot-js","Built for Node.js 18+, enabling server-side execution of the MCP Think Tank functionality.","01fkdhwcjc",{"_197":207,"_31":211,"_181":212},{"_199":208,"_33":209,"_202":210},"M1.857 18.013 11.736 24V12.456L1.857 6.468Zm20.286 0V6.468l-9.879 5.988V24Zm-.246-12.014L12 0 2.103 5.999 12 11.998Z","Task","task","Provides comprehensive task management tools for planning, tracking, and updating tasks with knowledge graph integration for persistent project management.","kyl8ik1uzw",{"_197":214,"_31":217,"_181":218},{"_199":215,"_33":175,"_202":216},"M1.125 0C.502 0 0 .502 0 1.125v21.75C0 23.498.502 24 1.125 24h21.75c.623 0 1.125-.502 1.125-1.125V1.125C24 .502 23.498 0 22.875 0zm17.363 9.75c.612 0 1.154.037 1.627.111a6.38 6.38 0 0 1 1.306.34v2.458a3.95 3.95 0 0 0-.643-.361 5.093 5.093 0 0 0-.717-.26 5.453 5.453 0 0 0-1.426-.2c-.3 0-.573.028-.819.086a2.1 2.1 0 0 0-.623.242c-.17.104-.3.229-.393.374a.888.888 0 0 0-.14.49c0 .196.053.373.156.529.104.156.252.304.443.444s.423.276.696.41c.273.135.582.274.926.416.47.197.892.407 1.266.628.374.222.695.473.963.753.268.279.472.598.614.957.142.359.214.776.214 1.253 0 .657-.125 1.21-.373 1.656a3.033 3.033 0 0 1-1.012 1.085 4.38 4.38 0 0 1-1.487.596c-.566.12-1.163.18-1.79.18a9.916 9.916 0 0 1-1.84-.164 5.544 5.544 0 0 1-1.512-.493v-2.63a5.033 5.033 0 0 0 3.237 1.2c.333 0 .624-.03.872-.09.249-.06.456-.144.623-.25.166-.108.29-.234.373-.38a1.023 1.023 0 0 0-.074-1.089 2.12 2.12 0 0 0-.537-.5 5.597 5.597 0 0 0-.807-.444 27.72 27.72 0 0 0-1.007-.436c-.918-.383-1.602-.852-2.053-1.405-.45-.553-.676-1.222-.676-2.005 0-.614.123-1.141.369-1.582.246-.441.58-.804 1.004-1.089a4.494 4.494 0 0 1 1.47-.629 7.536 7.536 0 0 1 1.77-.201zm-15.113.188h9.563v2.166H9.506v9.646H6.789v-9.646H3.375z","typescript","Built with TypeScript support, providing type safety for developers integrating with or extending the MCP Think Tank server.","cfpbo0ednc","namespace","npmPackage",{"_33":176,"_222":223},"weeklyDownloads",84,"prompts","redditPostForeignId","1khjez4","resources",[229],{"_230":231,"_33":232,"_233":234},"uri","status://health","Health","mimeType","text/plain","scores",{"_237":238,"_239":238,"_240":238},"license",100,"quality","security","simpleIcon","supportedPlatforms",[244,245,246],"WINDOWS","MACOS","LINUX","tools",[249,285,306,324,334,351,365,373,382,392,406,429,484,514,530,538,546,573,579,628],{"_33":250,"_31":251,"_252":253},"upsert_entities","Create new entities or update existing ones in the knowledge graph using an upsert pattern","inputSchema",{"_96":152,"_254":255,"_150":256,"_89":258,"_283":273},"$schema","http://json-schema.org/draft-07/schema#",[257],"entities",{"_257":259},{"_96":260,"_261":262,"_31":284},"array","items",{"_96":152,"_150":263,"_89":266,"_283":273},[33,264,265],"entityType","observations",{"_33":267,"_270":271,"_275":276,"_264":278,"_265":280},{"_96":97,"_268":184,"_31":269},"minLength","Unique identifier for the entity","update",{"_96":272,"_93":273,"_31":274},"boolean",false,"If true, will fully replace an existing entity; if false, will only create if entity doesn't exist","context",{"_96":97,"_31":277},"Optional context or situation relevant to this entity (e.g., project, meeting, or scenario)",{"_96":97,"_268":184,"_31":279},"Type classification of the entity",{"_96":260,"_261":281,"_31":282},{"_96":97},"Facts or observations about the entity","additionalProperties","Array of entities to create or update",{"_33":286,"_31":287,"_252":288},"create_relations","Create multiple new relations between entities in the knowledge graph. Relations should be in active voice",{"_96":152,"_254":255,"_150":289,"_89":291,"_283":273},[290],"relations",{"_290":292},{"_96":260,"_261":293,"_31":305},{"_96":152,"_150":294,"_89":298,"_283":273},[295,296,297],"from","to","relationType",{"_296":299,"_295":301,"_297":303},{"_96":97,"_268":184,"_31":300},"Target entity name",{"_96":97,"_268":184,"_31":302},"Source entity name",{"_96":97,"_268":184,"_31":304},"Type of relationship (in active voice)","Array of relations to create",{"_33":307,"_31":308,"_252":309},"add_observations","Add new observations to existing entities in the knowledge graph",{"_96":152,"_254":255,"_150":310,"_89":311,"_283":273},[265],{"_265":312},{"_96":260,"_261":313,"_31":323},{"_96":152,"_150":314,"_89":317,"_283":273},[315,316],"entityName","contents",{"_316":318,"_315":321},{"_96":260,"_261":319,"_31":320},{"_96":97},"Observations to add to the entity",{"_96":97,"_268":184,"_31":322},"Name of the entity to add observations to","Array of entity observations to add",{"_33":325,"_31":326,"_252":327},"delete_entities","Delete multiple entities and their associated relations from the knowledge graph",{"_96":152,"_254":255,"_150":328,"_89":330,"_283":273},[329],"entityNames",{"_329":331},{"_96":260,"_261":332,"_31":333},{"_96":97,"_268":184},"Array of entity names to delete",{"_33":335,"_31":336,"_252":337},"delete_observations","Delete specific observations from entities in the knowledge graph",{"_96":152,"_254":255,"_150":338,"_89":340,"_283":273},[339],"deletions",{"_339":341},{"_96":260,"_261":342,"_31":350},{"_96":152,"_150":343,"_89":344,"_283":273},[315,265],{"_315":345,"_265":347},{"_96":97,"_268":184,"_31":346},"Name of the entity to remove observations from",{"_96":260,"_261":348,"_31":349},{"_96":97},"Observations to remove from the entity","Array of entity observations to delete",{"_33":352,"_31":353,"_252":354},"delete_relations","Delete multiple relations from the knowledge graph",{"_96":152,"_254":255,"_150":355,"_89":356,"_283":273},[290],{"_290":357},{"_96":260,"_261":358,"_31":364},{"_96":152,"_150":359,"_89":360,"_283":273},[295,296,297],{"_296":361,"_295":362,"_297":363},{"_96":97,"_268":184,"_31":300},{"_96":97,"_268":184,"_31":302},{"_96":97,"_268":184,"_31":304},"Array of relations to delete",{"_33":366,"_31":367,"_252":368},"read_graph","Read the entire knowledge graph",{"_96":152,"_254":255,"_89":369,"_283":273},{"_370":371},"dummy",{"_96":97,"_31":372},"Placeholder parameter - this tool doesn't require parameters but returns the complete knowledge graph with entities and relationships",{"_33":374,"_31":375,"_252":376},"search_nodes","Search for nodes in the knowledge graph based on a query",{"_96":152,"_254":255,"_150":377,"_89":379,"_283":273},[378],"query",{"_378":380},{"_96":97,"_268":184,"_31":381},"Search query to find matching entities",{"_33":383,"_31":384,"_252":385},"open_nodes","Open specific nodes in the knowledge graph by their names",{"_96":152,"_254":255,"_150":386,"_89":388,"_283":273},[387],"names",{"_387":389},{"_96":260,"_261":390,"_31":391},{"_96":97,"_268":184},"Array of entity names to retrieve",{"_33":393,"_31":394,"_252":395},"update_relations","Update multiple existing relations in the knowledge graph",{"_96":152,"_254":255,"_150":396,"_89":397,"_283":273},[290],{"_290":398},{"_96":260,"_261":399,"_31":405},{"_96":152,"_150":400,"_89":401,"_283":273},[295,296,297],{"_296":402,"_295":403,"_297":404},{"_96":97,"_268":184,"_31":300},{"_96":97,"_268":184,"_31":302},{"_96":97,"_268":184,"_31":304},"Array of relations to update",{"_33":407,"_31":408,"_252":409},"memory_query","Query the memory store with advanced filters",{"_96":152,"_254":255,"_89":410,"_283":273},{"_38":411,"_413":414,"_416":417,"_419":420,"_423":424,"_426":427},{"_96":97,"_31":412},"Tag to filter observations by","after",{"_96":97,"_31":415},"ISO date to filter observations after","agent",{"_96":97,"_31":418},"Agent that created the observations","limit",{"_96":421,"_31":422},"number","Maximum number of results to return","before",{"_96":97,"_31":425},"ISO date to filter observations before","keyword",{"_96":97,"_31":428},"Text to search for in observations",{"_33":430,"_31":431,"_252":432},"think","Use the tool to think about something. It will not obtain new information or change the database, but just append the thought to the log. Use it when complex reasoning or some cache memory is needed. Consider including: problem definition, relevant context, analysis steps, self-reflection on your reasoning, and conclusions. Adapt this structure as needed for your specific thought process.",{"_96":152,"_254":255,"_150":433,"_89":435,"_283":273},[434],"structuredReasoning",{"_436":437,"_275":440,"_442":443,"_445":446,"_454":455,"_457":458,"_460":461,"_463":464,"_466":467,"_469":470,"_472":473,"_475":476,"_478":479,"_434":481},"tags",{"_96":260,"_261":438,"_31":439},{"_96":97},"Optional tags to help categorize and find this thought later",{"_96":97,"_31":441},"Optional context or situation relevant to this thought (e.g., project, meeting, or scenario)","category",{"_96":97,"_31":444},"Optional category for the thought (e.g., \"problem-solving\", \"analysis\", \"planning\")","formatType",{"_447":448,"_96":97,"_93":449,"_31":453},"enum",[449,450,451,452],"auto","general","problem","comparison","The type of formatting to apply","currentStep",{"_96":421,"_31":456},"The current step number in the thinking process","selfReflect",{"_96":272,"_93":273,"_31":459},"Whether to perform a self-reflection pass after generating the answer","formatOutput",{"_96":272,"_93":193,"_31":462},"Whether to apply markdown formatting to the output","plannedSteps",{"_96":421,"_31":465},"The total number of steps planned for this thinking process","allowResearch",{"_96":272,"_93":273,"_31":468},"Whether to allow research via external tools during the reasoning process","reflectPrompt",{"_96":97,"_31":471},"Custom prompt for the self-reflection stage","researchQuery",{"_96":97,"_31":474},"Optional research query to execute during the reasoning process","storeInMemory",{"_96":272,"_93":273,"_31":477},"Whether to store this thought in the knowledge graph memory","associateWithEntity",{"_96":97,"_31":480},"Optional entity name to associate this thought with",{"_96":97,"_268":482,"_31":483},10,"A structured thought process to work through complex problems. Use this as a dedicated space for reasoning step-by-step.",{"_33":485,"_31":486,"_252":487},"plan_tasks","Create multiple tasks from a plan. Generates IDs and syncs with knowledge graph.",{"_96":152,"_254":255,"_150":488,"_89":490,"_283":273},[489],"tasks",{"_489":491},{"_96":260,"_261":492,"_31":513},{"_96":152,"_150":493,"_89":494,"_283":273},[31],{"_495":496,"_436":499,"_501":502,"_507":508,"_31":511},"due",{"_96":97,"_497":498},"format","date-time",{"_96":260,"_261":500},{"_96":97},"priority",{"_447":503,"_96":97,"_93":505},[504,505,506],"low","medium","high","dependsOn",{"_96":260,"_261":509},{"_96":97,"_497":510},"uuid",{"_96":97,"_268":512},3,"List of tasks to create with their details",{"_33":515,"_31":516,"_252":517},"list_tasks","List tasks with optional filtering by status and priority.",{"_96":152,"_254":255,"_89":518,"_283":273},{"_519":520,"_501":527},"status",{"_447":521,"_96":97,"_31":526},[522,523,524,525],"todo","in-progress","blocked","done","Filter tasks by status",{"_447":528,"_96":97,"_31":529},[504,505,506],"Filter tasks by priority level",{"_33":531,"_31":532,"_252":533},"next_task","Get the next highest priority todo task and mark it as in-progress.",{"_96":152,"_254":255,"_89":534,"_283":273},{"_535":536},"random_string",{"_96":97,"_31":537},"Dummy parameter for no-parameter tools",{"_33":539,"_31":540,"_252":541},"complete_task","Mark a task as completed.",{"_96":152,"_254":255,"_150":542,"_89":543,"_283":273},[16],{"_16":544},{"_96":97,"_497":510,"_31":545},"UUID of the task to mark as completed",{"_33":547,"_31":548,"_252":549},"update_tasks","Update multiple tasks with new values.",{"_96":152,"_254":255,"_150":550,"_89":552,"_283":273},[551],"updates",{"_551":553},{"_96":260,"_261":554,"_31":572},{"_96":152,"_150":555,"_89":556,"_283":273},[16],{"_16":557,"_495":559,"_436":561,"_519":564,"_501":567,"_31":570},{"_96":97,"_497":510,"_31":558},"UUID of the task to update",{"_96":97,"_497":498,"_31":560},"New due date in ISO format",{"_96":260,"_261":562,"_31":563},{"_96":97},"New tags for the task",{"_447":565,"_96":97,"_31":566},[522,523,524,525],"New task status",{"_447":568,"_96":97,"_31":569},[504,505,506],"New priority level",{"_96":97,"_268":512,"_31":571},"New task description","List of task updates to apply",{"_33":574,"_31":575,"_252":576},"show_memory_path","Return absolute path of the active knowledge-graph file.",{"_96":152,"_254":255,"_89":577,"_283":273},{"_535":578},{"_96":97,"_31":537},{"_33":580,"_31":581,"_252":582},"exa_search","Search the web using Exa API",{"_96":152,"_254":255,"_150":583,"_89":584,"_283":273},[378],{"_96":585,"_378":589,"_442":591,"_603":604,"_609":610,"_614":615,"_618":619,"_622":623,"_625":626},{"_447":586,"_96":97,"_93":449,"_31":588},[449,426,587],"neural","Search type: auto (default), keyword (exact matching), or neural (semantic search)",{"_96":97,"_31":590},"The search query to execute",{"_447":592,"_96":97,"_93":450,"_31":602},[450,593,594,595,596,597,598,599,600,601],"company","research paper","news","pdf","github","tweet","personal site","linkedin profile","financial report","Filter results by content category","live_crawl",{"_447":605,"_96":97,"_93":606,"_31":608},[606,607],"always","fallback","When to use live crawling: 'always' or 'fallback' (when cached not available)","num_results",{"_96":421,"_93":189,"_611":238,"_612":184,"_31":613},"maximum","minimum","Number of results to return (1-100)","exclude_domains",{"_96":260,"_261":616,"_31":617},{"_96":97},"Exclude results from these domains","include_domains",{"_96":260,"_261":620,"_31":621},{"_96":97},"Only include results from these domains","end_published_date",{"_96":97,"_31":624},"Filter results published before this date (ISO format)","start_published_date",{"_96":97,"_31":627},"Filter results published after this date (ISO format)",{"_33":629,"_31":630,"_252":631},"exa_answer","Ask a question and get a sourced answer via Exa /answer API.",{"_96":152,"_254":255,"_150":632,"_89":634,"_283":273},[633],"question",{"_633":635,"_637":638},{"_96":97,"_268":189,"_31":636},"The question to ask and get an answer with sources","max_citations",{"_96":421,"_93":189,"_611":482,"_612":184,"_31":639},"Maximum number of citations to include in the answer (1-10)","pgg03wdx49","updatedAt","2025-05-26T17:05:15.261261705Z","meta",[645,648,650,657,660,662,665,668,671,674,677,679,682,685,687,689,691,693,695],{"_646":647},"title","MCP Think Tank | Glama",{"_649":42,"_33":31},"content",{"_651":652,"_653":654,"_655":656},"href","https://glama.ai/mcp/servers/@flight505/mcp-think-tank","rel","canonical","tagName","link",{"_649":43,"_658":659},"property","og:title",{"_649":42,"_658":661},"og:description",{"_649":663,"_658":664},"https://glama.ai/logo.png","og:logo",{"_649":666,"_658":667},"https://glama.ai/mcp/servers/@flight505/mcp-think-tank/og-image","og:image",{"_649":669,"_658":670},"630","og:image:height",{"_649":672,"_658":673},"1200","og:image:width",{"_649":675,"_658":676},"website","og:type",{"_649":652,"_658":678},"og:url",{"_649":680,"_658":681},"Glama – MCP Hosting Platform","og:site_name",{"_649":683,"_33":684},"summary_large_image","twitter:card",{"_649":680,"_33":686},"twitter:title",{"_649":42,"_33":688},"twitter:description",{"_649":666,"_33":690},"twitter:image",{"_649":669,"_33":692},"twitter:image:height",{"_649":672,"_33":694},"twitter:image:width",{"_696":697},"script:ld+json",{"_698":699,"_700":701},"@context","http://schema.org","@graph",[702],{"_698":703,"_704":705,"_706":707},"https://schema.org","@type","BreadcrumbList","itemListElement",[708,714,718,721],{"_704":709,"_710":711,"_33":712,"_713":184},"ListItem","item","https://glama.ai/","Glama","position",{"_704":709,"_710":715,"_33":716,"_713":717},"https://glama.ai/mcp","MCP",2,{"_704":709,"_710":719,"_33":720,"_713":512},"https://glama.ai/mcp/servers","Servers",{"_704":709,"_710":652,"_33":43,"_713":722},4,"visitorIsMaintainer","routes/_public/mcp/servers/~namespace/~slug/_pages/reviews/_route",{"_82":-5,"_29":726,"_643":959,"_993":-5},{"_85":86,"_87":727,"_13":745,"_161":42,"_162":42,"_163":746,"_167":747,"_190":-5,"_16":191,"_192":193,"_194":750,"_33":43,"_219":180,"_220":757,"_224":-5,"_225":226,"_227":758,"_235":760,"_241":-5,"_202":176,"_242":761,"_247":762,"_181":640,"_641":642},{"_89":728,"_150":744,"_96":152},{"_91":729,"_98":730,"_102":731,"_106":732,"_110":733,"_113":734,"_117":735,"_120":736,"_123":737,"_127":738,"_131":739,"_134":740,"_138":741,"_142":742,"_146":743},{"_93":94,"_31":95,"_96":97},{"_93":100,"_31":101,"_96":97},{"_93":104,"_31":105,"_96":97},{"_93":108,"_31":109,"_96":97},{"_31":112,"_96":97},{"_93":115,"_31":116,"_96":97},{"_93":94,"_31":119,"_96":97},{"_93":94,"_31":122,"_96":97},{"_93":125,"_31":126,"_96":97},{"_93":129,"_31":130,"_96":97},{"_93":94,"_31":133,"_96":97},{"_93":136,"_31":137,"_96":97},{"_93":140,"_31":141,"_96":97},{"_93":144,"_31":145,"_96":97},{"_93":148,"_31":149,"_96":97},[110],[154,155,156,157,158,159,160],{"_35":44,"_165":166},{"_169":170,"_171":172,"_173":-5,"_174":175,"_33":176,"_177":748,"_183":184,"_185":749,"_188":189},{"_179":180,"_181":182},{"_33":187},[751,753,755],{"_197":752,"_31":204,"_181":205},{"_199":200,"_33":201,"_202":203},{"_197":754,"_31":211,"_181":212},{"_199":208,"_33":209,"_202":210},{"_197":756,"_31":217,"_181":218},{"_199":215,"_33":175,"_202":216},{"_33":176,"_222":223},[759],{"_230":231,"_33":232,"_233":234},{"_237":238,"_239":238,"_240":238},[244,245,246],[763,777,788,799,805,816,827,831,836,842,853,862,882,898,905,909,914,931,935,953],{"_33":250,"_31":251,"_252":764},{"_96":152,"_254":255,"_150":765,"_89":766,"_283":273},[257],{"_257":767},{"_96":260,"_261":768,"_31":284},{"_96":152,"_150":769,"_89":770,"_283":273},[33,264,265],{"_33":771,"_270":772,"_275":773,"_264":774,"_265":775},{"_96":97,"_268":184,"_31":269},{"_96":272,"_93":273,"_31":274},{"_96":97,"_31":277},{"_96":97,"_268":184,"_31":279},{"_96":260,"_261":776,"_31":282},{"_96":97},{"_33":286,"_31":287,"_252":778},{"_96":152,"_254":255,"_150":779,"_89":780,"_283":273},[290],{"_290":781},{"_96":260,"_261":782,"_31":305},{"_96":152,"_150":783,"_89":784,"_283":273},[295,296,297],{"_296":785,"_295":786,"_297":787},{"_96":97,"_268":184,"_31":300},{"_96":97,"_268":184,"_31":302},{"_96":97,"_268":184,"_31":304},{"_33":307,"_31":308,"_252":789},{"_96":152,"_254":255,"_150":790,"_89":791,"_283":273},[265],{"_265":792},{"_96":260,"_261":793,"_31":323},{"_96":152,"_150":794,"_89":795,"_283":273},[315,316],{"_316":796,"_315":798},{"_96":260,"_261":797,"_31":320},{"_96":97},{"_96":97,"_268":184,"_31":322},{"_33":325,"_31":326,"_252":800},{"_96":152,"_254":255,"_150":801,"_89":802,"_283":273},[329],{"_329":803},{"_96":260,"_261":804,"_31":333},{"_96":97,"_268":184},{"_33":335,"_31":336,"_252":806},{"_96":152,"_254":255,"_150":807,"_89":808,"_283":273},[339],{"_339":809},{"_96":260,"_261":810,"_31":350},{"_96":152,"_150":811,"_89":812,"_283":273},[315,265],{"_315":813,"_265":814},{"_96":97,"_268":184,"_31":346},{"_96":260,"_261":815,"_31":349},{"_96":97},{"_33":352,"_31":353,"_252":817},{"_96":152,"_254":255,"_150":818,"_89":819,"_283":273},[290],{"_290":820},{"_96":260,"_261":821,"_31":364},{"_96":152,"_150":822,"_89":823,"_283":273},[295,296,297],{"_296":824,"_295":825,"_297":826},{"_96":97,"_268":184,"_31":300},{"_96":97,"_268":184,"_31":302},{"_96":97,"_268":184,"_31":304},{"_33":366,"_31":367,"_252":828},{"_96":152,"_254":255,"_89":829,"_283":273},{"_370":830},{"_96":97,"_31":372},{"_33":374,"_31":375,"_252":832},{"_96":152,"_254":255,"_150":833,"_89":834,"_283":273},[378],{"_378":835},{"_96":97,"_268":184,"_31":381},{"_33":383,"_31":384,"_252":837},{"_96":152,"_254":255,"_150":838,"_89":839,"_283":273},[387],{"_387":840},{"_96":260,"_261":841,"_31":391},{"_96":97,"_268":184},{"_33":393,"_31":394,"_252":843},{"_96":152,"_254":255,"_150":844,"_89":845,"_283":273},[290],{"_290":846},{"_96":260,"_261":847,"_31":405},{"_96":152,"_150":848,"_89":849,"_283":273},[295,296,297],{"_296":850,"_295":851,"_297":852},{"_96":97,"_268":184,"_31":300},{"_96":97,"_268":184,"_31":302},{"_96":97,"_268":184,"_31":304},{"_33":407,"_31":408,"_252":854},{"_96":152,"_254":255,"_89":855,"_283":273},{"_38":856,"_413":857,"_416":858,"_419":859,"_423":860,"_426":861},{"_96":97,"_31":412},{"_96":97,"_31":415},{"_96":97,"_31":418},{"_96":421,"_31":422},{"_96":97,"_31":425},{"_96":97,"_31":428},{"_33":430,"_31":431,"_252":863},{"_96":152,"_254":255,"_150":864,"_89":865,"_283":273},[434],{"_436":866,"_275":868,"_442":869,"_445":870,"_454":872,"_457":873,"_460":874,"_463":875,"_466":876,"_469":877,"_472":878,"_475":879,"_478":880,"_434":881},{"_96":260,"_261":867,"_31":439},{"_96":97},{"_96":97,"_31":441},{"_96":97,"_31":444},{"_447":871,"_96":97,"_93":449,"_31":453},[449,450,451,452],{"_96":421,"_31":456},{"_96":272,"_93":273,"_31":459},{"_96":272,"_93":193,"_31":462},{"_96":421,"_31":465},{"_96":272,"_93":273,"_31":468},{"_96":97,"_31":471},{"_96":97,"_31":474},{"_96":272,"_93":273,"_31":477},{"_96":97,"_31":480},{"_96":97,"_268":482,"_31":483},{"_33":485,"_31":486,"_252":883},{"_96":152,"_254":255,"_150":884,"_89":885,"_283":273},[489],{"_489":886},{"_96":260,"_261":887,"_31":513},{"_96":152,"_150":888,"_89":889,"_283":273},[31],{"_495":890,"_436":891,"_501":893,"_507":895,"_31":897},{"_96":97,"_497":498},{"_96":260,"_261":892},{"_96":97},{"_447":894,"_96":97,"_93":505},[504,505,506],{"_96":260,"_261":896},{"_96":97,"_497":510},{"_96":97,"_268":512},{"_33":515,"_31":516,"_252":899},{"_96":152,"_254":255,"_89":900,"_283":273},{"_519":901,"_501":903},{"_447":902,"_96":97,"_31":526},[522,523,524,525],{"_447":904,"_96":97,"_31":529},[504,505,506],{"_33":531,"_31":532,"_252":906},{"_96":152,"_254":255,"_89":907,"_283":273},{"_535":908},{"_96":97,"_31":537},{"_33":539,"_31":540,"_252":910},{"_96":152,"_254":255,"_150":911,"_89":912,"_283":273},[16],{"_16":913},{"_96":97,"_497":510,"_31":545},{"_33":547,"_31":548,"_252":915},{"_96":152,"_254":255,"_150":916,"_89":917,"_283":273},[551],{"_551":918},{"_96":260,"_261":919,"_31":572},{"_96":152,"_150":920,"_89":921,"_283":273},[16],{"_16":922,"_495":923,"_436":924,"_519":926,"_501":928,"_31":930},{"_96":97,"_497":510,"_31":558},{"_96":97,"_497":498,"_31":560},{"_96":260,"_261":925,"_31":563},{"_96":97},{"_447":927,"_96":97,"_31":566},[522,523,524,525],{"_447":929,"_96":97,"_31":569},[504,505,506],{"_96":97,"_268":512,"_31":571},{"_33":574,"_31":575,"_252":932},{"_96":152,"_254":255,"_89":933,"_283":273},{"_535":934},{"_96":97,"_31":537},{"_33":580,"_31":581,"_252":936},{"_96":152,"_254":255,"_150":937,"_89":938,"_283":273},[378],{"_96":939,"_378":941,"_442":942,"_603":944,"_609":946,"_614":947,"_618":949,"_622":951,"_625":952},{"_447":940,"_96":97,"_93":449,"_31":588},[449,426,587],{"_96":97,"_31":590},{"_447":943,"_96":97,"_93":450,"_31":602},[450,593,594,595,596,597,598,599,600,601],{"_447":945,"_96":97,"_93":606,"_31":608},[606,607],{"_96":421,"_93":189,"_611":238,"_612":184,"_31":613},{"_96":260,"_261":948,"_31":617},{"_96":97},{"_96":260,"_261":950,"_31":621},{"_96":97},{"_96":97,"_31":624},{"_96":97,"_31":627},{"_33":629,"_31":630,"_252":954},{"_96":152,"_254":255,"_150":955,"_89":956,"_283":273},[633],{"_633":957,"_637":958},{"_96":97,"_268":189,"_31":636},{"_96":421,"_93":189,"_611":482,"_612":184,"_31":639},[960,962,964,966,968,969,970,971,972,973,974,975,976,977,978,979,980,981,982],{"_646":961},"Reviews | MCP Think Tank | Glama",{"_649":963,"_33":31},"Read community reviews and feedback for MCP Think Tank. Discover user experiences, ratings, strengths, and potential limitations of this MCP implementation from actual users.",{"_651":965,"_653":654,"_655":656},"https://glama.ai/mcp/servers/@flight505/mcp-think-tank/reviews",{"_649":967,"_658":659},"Reviews | MCP Think Tank",{"_649":963,"_658":661},{"_649":663,"_658":664},{"_649":666,"_658":667},{"_649":669,"_658":670},{"_649":672,"_658":673},{"_649":675,"_658":676},{"_649":965,"_658":678},{"_649":680,"_658":681},{"_649":683,"_33":684},{"_649":680,"_33":686},{"_649":963,"_33":688},{"_649":666,"_33":690},{"_649":669,"_33":692},{"_649":672,"_33":694},{"_696":983},{"_698":699,"_700":984},[985],{"_698":703,"_704":705,"_706":986},[987,988,989,990,991],{"_704":709,"_710":711,"_33":712,"_713":184},{"_704":709,"_710":715,"_33":716,"_713":717},{"_704":709,"_710":719,"_33":720,"_713":512},{"_704":709,"_710":652,"_33":43,"_713":722},{"_704":709,"_710":965,"_33":992,"_713":189},"Reviews","review","actionData","errors"]

Rate & Review

MCP directory API