@pieeee/docmcp

DocMCP

Index any documentation website and search it from AI coding assistants via the Model Context Protocol (MCP).

Features

• Crawl & Index: Automatically crawl documentation sites via sitemap or recursive links
• Hybrid Search: Combines BM25 keyword search with vector embeddings for best results
• MCP Integration: Works with Claude Code, Claude Desktop, Cursor, and any MCP-compatible tool
• Multiple Providers: Anthropic (Voyage), OpenAI, or BM25-only (zero setup)
• Cross-Platform: Works on macOS, Linux, and Windows

Installation

npm install -g @pieeee/docmcp

Requirements

• Node.js 20+
• One of: Anthropic API key, OpenAI API key, or use BM25-only mode (no API needed)

Quick Start

# Initial setup
docmcp init

# Index a documentation site
docmcp add https://tailwindcss.com/docs

# List indexed docs
docmcp list

MCP Configuration

Claude Code

claude mcp add docmcp -- docmcp serve

Or add to your project's .mcp.json:

{
  "mcpServers": {
    "docmcp": {
      "command": "docmcp",
      "args": ["serve"]
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "docmcp": {
      "command": "docmcp",
      "args": ["serve"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "docmcp": {
      "command": "docmcp",
      "args": ["serve"]
    }
  }
}

CLI Commands

Command	Description
docmcp init	Setup wizard - configure embedding provider and data directory
docmcp add <url>	Crawl and index a documentation site
docmcp list	Show all indexed documentation
docmcp remove <name>	Remove indexed documentation
docmcp serve	Start MCP server (stdio transport)

Add Command Options

docmcp add <url> [options]

Options:
  -n, --name <name>           Override auto-detected doc name
  -d, --depth <number>        Max crawl depth (default: 10)
  -m, --max-pages <number>    Max pages to crawl (default: unlimited)
  -i, --include <pattern...>  Only crawl URLs matching pattern (glob)
  -e, --exclude <pattern...>  Skip URLs matching pattern (glob)
  --delay <ms>                Delay between requests (default: 200)
  --concurrency <number>      Parallel requests (default: 3)
  --no-sitemap                Skip sitemap, force recursive crawl

MCP Tools

When connected as an MCP server, DocMCP exposes these tools:

Tool	Description
search_docs	Search indexed documentation with hybrid BM25 + vector search
list_docs	List all indexed documentation sources

search_docs

Search your indexed documentation:

search_docs(query: "how to center a div", doc?: "Tailwind", limit?: 5)

Parameters:

• query (required): Search query
• doc (optional): Filter to specific documentation
• limit (optional): Max results (default: 5)

Embedding Providers

Provider	API Key Required	Notes
anthropic	ANTHROPIC_API_KEY	Uses Voyage AI embeddings (recommended)
openai	OPENAI_API_KEY	Uses text-embedding-3-small
bm25only	None	Keyword search only, zero setup

Set your API key as an environment variable or enter it during docmcp init.

Data Storage

All data is stored in ~/.docmcp/:

~/.docmcp/
├── config.json    # Configuration (API keys stored here)
└── db/
    └── docs.db    # SQLite database with FTS5 + vector search

Platform Support

Platform	Status	Notes
macOS (Intel)	Full
macOS (Apple Silicon)	Full
Linux (x64)	Full
Linux (ARM64)	Full
Windows (x64)	Full	May require build tools for native modules

Windows Prerequisites

If installation fails on Windows due to native module compilation:

• Install Visual Studio Build Tools
• Or run: npm install --global windows-build-tools
• Retry: npm install -g docmcp

How It Works

• Crawl: DocMCP crawls documentation sites using sitemap or recursive link following
• Parse: HTML is cleaned and converted to Markdown, preserving code blocks
• Chunk: Content is split at heading boundaries into ~512 token chunks
• Index: Chunks are stored in SQLite with FTS5 (BM25) and vector embeddings
• Search: Queries use hybrid search combining keyword and semantic matching

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT - see LICENSE for details.