MCP AI Bug Helper
Get multi-model coding assistance from three AI advisors running in parallel, with automatic fallback handling and intelligent model selection. Query coding specialists via the ask-coding-advisors tool with a question (4-4000 characters) and optional context (up to 12,000 characters for stacktraces, logs, or diffs). The system starts with free-tier models (qwen/qwen3-coder:free, qwen/qwen-2.5-coder-32b-instruct:free, agentica-org/deepcoder-14b-preview:free) and automatically switches to premium models (Claude 3.7 Sonnet, Codestral, GPT-5.1 Codex) when encountering rate limits. Features include scenario-based model selection using detected tags (React, Go, Node, infrastructure, LLM), health monitoring that temporarily disables failing models, detailed telemetry with token usage and latency statistics, and zero-installation deployment via npx @meinzeug/mcp-ai-bug-helper with MCP clients like Codex CLI, Claude Code, VS Code, and Cursor.
Uses dotenv for secure management of API keys and configuration through environment variables loaded from .env files.
Source code hosted on GitHub at meinzeug/mcp-ai-bug-helper for local development and contributions.
Compatible with JetBrains AI Assistant for integrating the coding advisor tools within JetBrains IDEs.
Uses Mermaid diagrams in documentation to visualize the advisor pipeline sequence between Codex, MCP Server, and OpenRouter.
Requires Node.js ≥20.10 as the runtime environment for executing the MCP server.
Distributed as a scoped npm package (@meinzeug/mcp-ai-bug-helper) that can be installed and run via npx for instant deployment.
Supports scenario-based model selection with specific optimizations for React-related coding questions and frontend development issues.
Displays project status badges for npm version, downloads, Node requirements, MCP compatibility, and license information.
References TOML configuration format for Windows 11 Codex integration setup in ~/.codex/config.toml.
Utilized in development mode to run the TypeScript-based MCP server with real-time logging.
Implemented in TypeScript with full type definitions, sourcemaps, and type checking capabilities.
Listed as a compatible MCP client that can integrate the coding advisor functionality.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@MCP AI Bug HelperWhy does my React component re-render infinitely when using useEffect?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
🔗 Inhaltsverzeichnis
Related MCP server: Ask MCP
✨ Highlights
🧠 Mehrstufige Advisor-Pipeline: 3 kostenlose OpenRouter-Codingmodelle + automatische Premium-Fallbacks bei Rate-Limits.
🤖 Codex-native Instruktionen: MCP Instructions erklären dem Agent genau, wann das Tool sinnvoll ist.
📉 Kosten- & Latenz-Telemetrie: Jede Antwort liefert Token-Usage + Antwortzeit direkt an Codex zurück.
⚡
npx-fähig:npx @meinzeug/mcp-ai-bug-helperreicht – kein Clonen erforderlich.🔐 Secretsafe:
.env+dotenvfür API Keys, keine versehentlichen Commits.
⚡️ Schnellstart
Option A – Zero Install via npx
export OPENROUTER_API_KEY=sk-or-v1-... # einmalig setzen
npx @meinzeug/mcp-ai-bug-helperOption B – Lokal entwickeln
git clone https://github.com/meinzeug/mcp-ai-bug-helper.git
cd mcp-ai-bug-helper
npm install
cp .env.example .env && $EDITOR .env # Key eintragen
npm run dev💡
npm run devstartet den MCP-Server viats-nodeund streamt Logs – perfekt zum Debuggen.
📦 Veröffentlichtes Paket:
@meinzeug/mcp-ai-bug-helper– überprüfbar mitnpm view @meinzeug/mcp-ai-bug-helper.
🛠️ Tool-Flow
Phase | Beschreibung |
🔍 Analyse |
|
🧑💻 Advisor-Fächer |
|
🚨 Fallback-Trigger | Erstes HTTP 429 schaltet auf Premium-Rotation: |
📊 Response-Payload | Drei unabhängige Antworten + Latency + Token-Stats + Hinweis, ob Fallback aktiv war. |
Adaptive Modellwahl
🔎 Jeder Prompt wird analysiert (Keywords für React/Go/infra/LLM etc.), daraus entstehen
ScenarioTags(z. B.frontend,node,go).🤖 Für jeden Tag gibt es passende Modelle mit hinterlegten Stärken und Zuverlässigkeitsskalen (platinum/gold/silver).
✅ Vor jedem Call prüft der Server über
GET /api/v1/models, ob das Modell beim Account verfügbar ist. Modelle mit 404/500 werden für einige Minuten automatisch gesperrt.💳 Wenn kein gesundes Free-Modell verfügbar ist oder ein 429 zurückkommt, wird automatisch auf Premium-Fallbacks (Claude, Codestral, GPT-5.1 Codex) gewechselt.
🚫 Sobald die kostenlosen Modelle dauerhaft fehlschlagen (z. B. Quota aufgebraucht oder Provider-404), werden sie für den gesamten Prozess deaktiviert – alle weiteren Aufrufe laufen direkt über die kostenpflichtigen Advisors.
sequenceDiagram
participant Codex
participant MCP as MCP Server
participant OR as OpenRouter
Codex->>MCP: call ask-coding-advisors(question, context)
loop 3 advisors
MCP->>OR: chat(model=free[i])
alt HTTP 429
MCP->>OR: chat(model=paid[j])
end
OR-->>MCP: text + usage + latency
MCP-->>Codex: stream advisor result
end🤝 Codex-Integration
Server in Codex registrieren
codex mcp add coding-advisors \ --env "OPENROUTER_API_KEY=sk-or-v1-..." \ npx @meinzeug/mcp-ai-bug-helper # Syntax entspricht auch anderen MCP-Beispielen wie # codex mcp add chrome-devtools -- npx chrome-devtools-mcp@latestVerifizieren
codex mcp list | grep coding-advisorsIm Prompt nutzen
/tools→ask-coding-advisorsoder direkt mit
@ask-coding-advisorsim Prompt.
📎 Wenn du lieber aus dem Repo startest:
--cmd "node" --args "dist/server.js"verwenden.
Andere MCP-Clients
Claude Code: Anthropic beschreibt in den Claude-Code-MCP-Docs exakt den gleichen stdio-/HTTP-Mechanismus. Du kannst denselben Befehl wie oben benutzen (oder einen
claude.json-Eintrag), und die Tools erscheinen dort unter/mcpbzw.@ask-coding-advisors.Sonstige IDEs: Jede Umgebung, die MCP spricht (z. B. VS Code-Extensions, Cursor, Eigene Agenten), kann denselben Server starten. Wichtig ist nur, dass
OPENROUTER_API_KEYgesetzt ist und der Prozess via stdio oder TCP erreichbar ist.
Schnellbefehle für gängige Clients
Amp CLI –
amp mcp add coding-advisors -- npx @meinzeug/mcp-ai-bug-helperClaude Code CLI –
claude mcp add coding-advisors npx @meinzeug/mcp-ai-bug-helperCodex CLI –
codex mcp add coding-advisors --env "OPENROUTER_API_KEY=sk-or-v1-..." npx @meinzeug/mcp-ai-bug-helperCopilot CLI –
/mcp add→ Namecoding-advisors, Commandnpx @meinzeug/mcp-ai-bug-helperVS Code / Copilot –
code --add-mcp '{"name":"coding-advisors","command":"npx","args":["@meinzeug/mcp-ai-bug-helper"]}'Cursor / Amp / Cline / Kiro / Qoder / Warp / Windsurf / JetBrains AI / Gemini / Droid (Factory CLI) – überall einfach denselben Command + Name übernehmen; falls der Client eine Konfigurationsdatei verlangt, setzt du
command = "npx",args = ["@meinzeug/mcp-ai-bug-helper"]und ergänztOPENROUTER_API_KEYin der jeweiligenenv-Sektion. Mobile Beispiele findest du in deren MCP-Handbüchern (Links: Amp, Cline, Gemini CLI, JetBrains AI Assistant).Windows 11 & Codex – falls
npxChrome oder Node ausProgram Filesladen muss, ergänze in~/.codex/config.toml:[mcp_servers.coding-advisors] command = "cmd" args = ["/c", "npx", "@meinzeug/mcp-ai-bug-helper"] env = { SystemRoot="C:\\Windows", PROGRAMFILES="C:\\Program Files", OPENROUTER_API_KEY="sk-or-v1-..." } startup_timeout_ms = 20_000
⚙️ Konfiguration
Variable | Pflicht | Beschreibung |
| ✅ | Persönlicher OpenRouter Key (liegt in |
| ⛔ optional | Eigener Analytics-Tag (Default: |
| ⛔ optional | Referrer-URL für OpenRouter-Dashboard. |
.env.example liefert ein Template. Für Mehrfach-Workspaces einfach mehrere .env Dateien pflegen und vor dem Start sourcen.
🏗️ Architektur
Transport:
@modelcontextprotocol/sdk+StdioServerTransportDomainlogik:
CodingAdvisorCoordinatororchestriert freie + paid Modelle.HTTP-Layer:
OpenRouterClient(nativefetch, Retry auf Rate-Limits, Usage-Mapping).Config:
config.tsliest.env,assertConfig()schützt vor fehlendem Key.Packaging: Scoped npm Modul,
bin→dist/server.js,prepare/postbuilderzeugen ausführbares Artefakt.
src/
├─ server.ts # MCP Bootstrap + Tool Definition
├─ codingAdvisors.ts # Sequencer + Fallback-Rotation
├─ openrouterClient.ts # REST Wrapper + Error Handling
├─ modelCatalog.ts # Liste freier & paid Modelle
├─ config.ts # dotenv + Validation
└─ errors.ts / types.ts # Hilfsklassen & DTOs🧪 Entwicklungs-Workflow
Script | Zweck |
| MCP Server mit |
| TypeScript Build + |
| Schneller TS-Check ohne Emit. |
| Führt das gebaute CLI ( |
| Führt reale OpenRouter-Tests auf mehreren Debug-Szenarien aus (filterbar via |
Bei Veröffentlichung sorgt
npm publishautomatisch für frische Builds (viaprepare).
🧩 Troubleshooting
Missing OPENROUTER_API_KEY–.envnicht geladen? Terminal neu starten odersource .env.429 Too Many Requests– Die App schwenkt automatisch auf die Premiumliste. Wenn alles blockiert ist, hilft nur Warten oder eigener OpenRouter-Plan.codex mcp add ...schlägt fehl – Prüfe, obcodexZugriff aufnpxhat (Pfad) oder wechsle auf direktennode dist/server.jsBefehl.Keine Antworten im Codex-UI –
npm run devseparat starten und schauen, ob Requests ankommen (stdout).
📜 Lizenz
MIT © meinzeug – Mit Liebe für MCP + Codex gebaut. Contributions willkommen! 🎉
Available Tools
1 toolask-coding-advisorsAsk Coding AdvisorsB
Queries three OpenRouter coding specialists (free tier) and falls back to premium models when rate limited.
| Name | Required | Description | Default |
|---|---|---|---|
| question | Yes | ||
| context | No |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden of behavioral disclosure. It adds useful context: it queries three specialists (implying multiple responses or aggregation), uses free tier models, and has a fallback mechanism for rate limits. However, it doesn't cover other key behaviors like response format, error handling beyond rate limits, or whether it's read-only or mutative, leaving gaps in transparency.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, efficient sentence that front-loads the core action ('Queries three OpenRouter coding specialists') and adds necessary details (free tier, fallback) without waste. Every part earns its place, making it highly concise and well-structured.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the complexity (querying multiple models with fallback), no annotations, no output schema, and 0% schema description coverage, the description is incomplete. It lacks details on parameter usage, response format, error handling beyond rate limits, and other operational aspects, making it inadequate for full contextual understanding despite its conciseness.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema has 2 parameters with 0% description coverage, so the description must compensate. It provides no information about the parameters (e.g., what 'question' and 'context' represent, their expected content, or how they're used in the query). This fails to add meaning beyond the bare schema, resulting in a low score due to the high burden from lack of schema documentation.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose: it queries three OpenRouter coding specialists. It specifies the service (OpenRouter), the type of specialists (coding), and the quantity (three). However, it doesn't distinguish from siblings since there are none, so it can't achieve a perfect score for sibling differentiation.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage context by mentioning 'free tier' and 'falls back to premium models when rate limited,' suggesting it's suitable for general coding queries with fallback handling. However, it lacks explicit guidance on when to use this tool versus alternatives (e.g., other query tools or direct model calls), and there are no siblings to compare against, so the guidance is limited to implied context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
1 tool update
v1.0.0- Changed
ask-coding-advisors2 fields changed- added
Input schema / $schemaAdded value: +"http://json-schema.org/draft-07/schema#" - added
Input schema / additionalPropertiesAdded value: +false
1 tool update
- First observed
ask-coding-advisors
TDQS
With only one tool, there is no possibility of ambiguity or overlap between tools. The tool has a distinct purpose of querying coding specialists, so an agent cannot misselect between non-existent alternatives.
A single tool inherently has perfect naming consistency, as there are no other tools to compare it against. The tool name 'ask-coding-advisors' follows a clear verb_noun pattern, but consistency cannot be assessed across a set of one.
One tool is generally too few for a server's purpose, as it limits functionality and suggests an incomplete or trivial scope. For a bug helper server, a single tool is insufficient to cover typical operations like analyzing, fixing, or testing code, making it feel thin and under-scoped.
The server's purpose as a bug helper implies a need for tools to handle various aspects of debugging, but with only one tool for querying advisors, there are significant gaps. Missing operations like code analysis, error detection, or solution implementation make the surface severely incomplete for the domain.
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
Multi-model code review: a panel of models + detectors return a pass/fail verdict. Paid via x402.
MCP-native AI SRE: ask what's broken in production, get a reviewed GitHub fix PR.
Coordinate coding agents through MCP using existing AI plans, saved work, and independent checks.
Multi-LLM council: 25+ frontier models in parallel, consensus scoring, verdict-first code review.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceProvides AI-powered assistance for coding problems using Google's Gemini AI, combined with Perplexity insights and Stack Overflow references, facilitating contextual analysis and automatic response archiving for improved troubleshooting.15-
- AlicenseNot gradedqualityDmaintenanceProvides access to multiple AI models (Grok, Gemini 2.5 Pro, Kimi, Qwen3 Coder, GLM-4.5) through OpenRouter, enabling users to query different specialized models for reasoning, coding, translation, and general tasks with automatic fallback to free variants.8MIT
- AlicenseBqualityDmaintenanceEnables AI agents to consult expert models (Claude, GPT, Gemini, DeepSeek, Z.ai) for technical guidance, code reviews, and architectural advice without switching context.4114MIT
- AlicenseAqualityDmaintenanceEnables AI assistants to consult other AI models via OpenRouter for help with coding tasks.190MIT
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/meinzeug/mcp-ai-bug-helper'
If you have feedback or need assistance with the MCP directory API, please join our Discord server