Documentation MCP

An MCP (Multi-modal Communication Protocol) server for Claude and other LLMs that fetches, parses, stores, and serves documentation. This server allows LLMs to access documentation in a structured format, similar to how Cursor Editor provides context to code assistants.

Features

• Fetch Documentation: Retrieve HTML content from documentation websites
• Parse Content: Convert HTML to Markdown for better LLM processing
• Local Storage: Store all documentation locally in SQLite
• Vector Search: Optional vector storage for semantic search
• WebSocket API: Real-time communication with LLMs
• REST API: HTTP endpoints for integration with applications
• CLI Interface: Easy to use command-line interface

Installation

You can use this MCP server directly with npx:

npx @anishpras/doc-mcp

Or install it globally:

npm install -g @anishpras/doc-mcp
doc-mcp

Usage

Command Line Options

Usage: doc-mcp [options]

Documentation MCP server for fetching and storing documentation for LLMs

Options:
  -V, --version           output the version number
  -p, --port <number>     Port to run the server on (default: auto-detect free port)
  -d, --data <path>       Path to store data (default: ./doc-mcp-data)
  -v, --vectors           Enable vector storage for semantic search (default: false)
  --dimension <number>    Vector dimension for embeddings (default: 1536)
  --verbose               Enable verbose logging (default: false)
  -h, --help              display help for command

Integrating with Claude Desktop

To use this MCP with Claude Desktop or other LLM applications, add the following configuration to your MCP settings:

{
  "documentation-mcp": {
    "command": "npx",
    "args": ["@anishpras/doc-mcp"]
  }
}

MCP Protocol

The Documentation MCP supports the following message types:

Fetching Documentation

Fetch and store documentation from a URL:

{
  "type": "fetch_documentation",
  "url": "https://example.com/docs",
  "selector": "main", // Optional CSS selector
  "forceRefresh": false // Optional, set to true to re-fetch
}

Response:

{
  "type": "documentation_fetched",
  "status": "success",
  "message": "Documentation fetched and stored successfully",
  "url": "https://example.com/docs",
  "documentId": 1,
  "title": "Example Documentation",
  "sectionCount": 10
}

Getting Documentation

Retrieve stored documentation:

{
  "type": "get_documentation",
  "url": "https://example.com/docs"
  // alternatively, you can use documentId
  // "documentId": 1
}

Response:

{
  "type": "documentation",
  "documentId": 1,
  "title": "Example Documentation",
  "content": "Full markdown content...",
  "chunks": [
    {
      "index": 0,
      "content": "Section content..."
    },
    // More chunks...
  ]
}

Searching Documentation

Search for documentation by keyword:

{
  "type": "search_documentation",
  "query": "api authentication",
  "limit": 10, // Optional
  "semantic": false // Optional, set to true for vector search
}

Response:

{
  "type": "search_results",
  "query": "api authentication",
  "results": [
    {
      "id": 1,
      "title": "Authentication Guide",
      "path": "",
      "url": "https://example.com/docs/auth",
      "snippet": "API authentication requires..."
    },
    // More results...
  ]
}

Listing Documentation

List all available documentation:

{
  "type": "list_documentation",
  "limit": 20, // Optional
  "offset": 0 // Optional
}

Response:

{
  "type": "documentation_list",
  "documents": [
    {
      "id": 1,
      "title": "Example Documentation",
      "url": "https://example.com/docs",
      "path": "",
      "updated_at": "2023-01-01T00:00:00.000Z",
      "chunk_count": 10
    },
    // More documents...
  ],
  "total": 50,
  "limit": 20,
  "offset": 0
}

REST API

The server also provides a REST API for HTTP-based integrations:

• GET /health - Check if the server is running
• POST /api/fetch - Fetch documentation from a URL
• GET /api/docs - Retrieve stored documentation
• GET /api/search - Search for documentation

Example Client Usage

Here's how to connect to the MCP from a client application:

const socket = new WebSocket('ws://localhost:3000');

// Listen for messages
socket.addEventListener('message', (event) => {
  const response = JSON.parse(event.data);
  console.log('Received response:', response);
});

// Connect and send a message
socket.addEventListener('open', () => {
  // Fetch documentation
  socket.send(JSON.stringify({
    type: 'fetch_documentation',
    url: 'https://docs.example.com/api'
  }));
});

Development

If you want to modify or contribute to this project:

1. Clone the repository
2. Install dependencies: npm install
3. Build the TypeScript code: npm run build
4. Run the development server: npm run dev

License

MIT

doc-mcp