Eine Befehlszeilen-Chat-Schnittstelle, die über die Anthropic API eine Verbindung zu einem MCP-Server herstellt. Erstellt während der Arbeit am Kurs Introduction to Model Context Protocol von Anthropic, erweitert um benutzerdefinierte Werkzeuge, Ressourcen und Prompts.
MCP (Model Context Protocol) ist ein offener Standard zur Verbindung von KI-Modellen mit externen Werkzeugen und Datenquellen. Dieses Projekt implementiert beide Seiten dieser Verbindung – einen FastMCP-Server, der Dokumente als Ressourcen bereitstellt und Werkzeuge zu deren Lesen und Bearbeiten definiert, sowie einen Client, der eine Verbindung zum Server herstellt und diese Funktionen innerhalb einer Chat-Schnittstelle verfügbar macht.
Der Server definiert:
Werkzeuge — Dokumente lesen und bearbeiten
Ressourcen — Alle Dokumente auflisten oder ein spezifisches Dokument per URI abrufen
Prompts — Ein Dokument in Markdown umformatieren oder dessen Inhalt zusammenfassen
Der Client implementiert die vollständige MCP-Client-Sitzung, einschließlich Werkzeugaufrufen, Ressourcenlesen, Prompt-Abruf und Befehls-Autovervollständigung.
Related MCP server: MCP Chat
Ausgehend von einem Kurs-Starterpaket habe ich die fehlenden Teile auf beiden Seiten implementiert:
read_resource, list_prompts und get_prompt auf der Client-Seite
Ressourcen-Endpunkte (docs://documents und docs://documents/{doc_id}) auf der Server-Seite
Zwei Prompts (format und summarize), die das Modell anweisen, die verfügbaren Werkzeuge zu nutzen
Das Wichtigste, was mir dieses Projekt verdeutlicht hat, ist die Trennung zwischen dem Server (der definiert, was verfügbar ist) und dem Client (der weiß, wie man es aufruft) – und dass Prompts lediglich strukturierte Nachrichten sind, die dem Modell einen Startkontext geben, keine Magie.
Python 3.9+
Anthropic API-Schlüssel
Klonen Sie das Repo und navigieren Sie in den Projektordner.
Erstellen Sie eine virtuelle Umgebung und aktivieren Sie diese:
uv venv
.venv\Scripts\activate # Windows
source .venv/bin/activate # Mac/LinuxInstallieren Sie die Abhängigkeiten:
uv pip install -e .Erstellen Sie eine .env-Datei im Projektstammverzeichnis:
ANTHROPIC_API_KEY="your-key-here"Starten Sie die App:
uv run main.pyGeben Sie eine Nachricht ein, um zu chatten. Verwenden Sie @doc_id, um ein Dokument in Ihre Anfrage einzubeziehen, und /command, um einen Prompt auszulösen. Die Tab-Taste vervollständigt verfügbare Befehle automatisch.
> Tell me about @deposition.md
> /summarize report.pdf
> /format plan.mdUm eigene Dokumente hinzuzufügen, bearbeiten Sie das docs-Dictionary in mcp_server.py.
mcp dev mcp_server.pyDies öffnet den MCP Inspector in Ihrem Browser, wo Sie Werkzeuge, Ressourcen und Prompts ohne die Chat-Schnittstelle testen können.
mcp_chat_cli/
├── main.py # entrypoint
├── mcp_server.py # FastMCP server — tools, resources, prompts
├── mcp_client.py # MCP client session wrapper
├── core/
│ ├── chat.py # chat loop logic
│ ├── claude.py # Anthropic API integration
│ ├── cli.py # CLI setup and input handling
│ ├── cli_chat.py # connects CLI and chat
│ └── tools.py # tool call handling
├── .env # API key (not committed)
└── pyproject.toml # dependenciesEdit a document by replacing a string in the documents content with a new string.
| Name | Required | Description | Default |
|---|---|---|---|
| doc_id | Yes | Id of the document that will be edited | |
| old_str | Yes | The text to replace. Must match exactly, including whitespace. | |
| new_str | Yes | The new text to insert in place of the old text. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are present, so the description must disclose behavioral traits. It only states 'replacing a string' without indicating whether it replaces all occurrences, behavior on missing string, or side effects like auto-save. Missing important mutation details.
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 sentence that is concise and front-loaded with the essential action and resource. No wasted words.
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?
For a simple edit tool with fully documented parameters, the description is adequate but lacks details on error handling (e.g., what if old_str not found) and whether it replaces all occurrences or just the first.
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?
Schema description coverage is 100%, and the description adds value by clarifying that old_str must match exactly including whitespace. This goes beyond the schema's basic descriptions.
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 action (edit), resource (document), and method (string replacement). It effectively distinguishes from the sibling tool 'read_doc_contents' which is read-only.
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 does not explicitly state when to use this tool vs alternatives. While the sibling tool name implies a read vs. edit distinction, no further guidance is provided on prerequisites or exclusions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Read the contents of a document and return it as a string.
| Name | Required | Description | Default |
|---|---|---|---|
| doc_id | Yes | Id of the document to read |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the burden of behavioral disclosure. It does state the output behavior ('return it as a string'), but does not mention error handling, permissions, or what happens when the document does not exist. For a simple read operation, this is adequate but not rich.
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, front-loaded sentence with no wasted words. It efficiently conveys both the operation and the return format.
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?
This is a simple tool with one parameter, full schema coverage, and no output schema. The description is largely complete for a read operation, especially since it explicitly states the return type. It could be slightly stronger with error or permission context, but nothing essential is missing for basic invocation.
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 schema already fully documents the only parameter, doc_id, as 'Id of the document to read', so schema coverage is 100%. The description adds no additional parameter meaning beyond the schema, matching the baseline score.
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 states a specific verb and resource: 'Read the contents of a document' and explicitly notes the return type as a string. This clearly distinguishes it from the sibling 'edit_document', which implies modification rather than reading.
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 verb 'Read' establishes a clear context for when this tool is appropriate, and its contrast with 'edit_document' implies read-only usage. However, it does not explicitly state when not to use it or name the alternative.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0edit_documentread_doc_contentsScored across 2 tools
The two tools have completely distinct purposes: one reads document contents, the other edits by replacing text. There is no functional overlap.
Both tool names follow the verb_noun pattern (edit_document, read_doc_contents), with consistent snake_case and clear action-object structure.
With only two tools, the server feels thin for a document utility, though it could be a minimal implementation. It falls into the borderline category (1-2 tools).
The server lacks basic CRUD operations like create and delete documents. While read and edit are provided, agents cannot create new documents or remove them, which are significant gaps.
Search your AI chat history (ChatGPT, Claude, Codex) from any MCP client. Remote, private, read-only
Build your own tools from chat: design, test and host them on a shareable MCP server URL.
Augments MCP Server - A comprehensive framework documentation provider for Claude Code
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
