docpull-mcp

MCP server for fetching and searching documentation on-demand. Uses docpull to pull docs and pgvector for semantic search.

Features

• Fetch docs on-demand - Pull documentation for any library
• Semantic search - Find concepts even when you don't know exact terms
• Exact matching - Grep-like search for known function/method names
• Built-in sources - React, Next.js, Hono, Supabase, and more
• Custom sources - Add your own documentation URLs
• 7-day cache - Automatic refresh of stale docs

Install

git clone https://github.com/raintree-technology/docpull-mcp
cd docpull-mcp
bun install

# Requires docpull CLI
pip install docpull

For semantic search (optional but recommended)

# PostgreSQL with pgvector
psql $DATABASE_URL -f schema.sql

# Set environment variables
export DATABASE_URL="postgresql://user:pass@localhost:5432/docs"
export OPENAI_API_KEY="sk-..."

Usage

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "docpull": {
      "command": "bun",
      "args": ["run", "/path/to/docpull-mcp/src/server.ts"],
      "env": {
        "DATABASE_URL": "postgresql://...",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "docpull": {
      "command": "bun",
      "args": ["run", "/path/to/docpull-mcp/src/server.ts"],
      "env": {
        "DATABASE_URL": "postgresql://...",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Tools

ensure_docs

Fetch and index documentation for a library.

ensure_docs(source: "react")              # Fetch and index
ensure_docs(source: "https://...")        # Direct URL
ensure_docs(source: "react", force: true) # Force refresh
ensure_docs(source: "react", index: false) # Fetch only, no indexing

search_docs

Semantic search for concepts (requires DATABASE_URL + OPENAI_API_KEY).

search_docs(query: "how to stream responses", library: "anthropic")
search_docs(query: "row level security", library: "supabase")

grep_docs

Fast exact pattern matching (requires DATABASE_URL).

grep_docs(pattern: "onConflictDoUpdate", library: "drizzle")
grep_docs(pattern: "usePrefetchQuery")

list_sources

List available documentation sources.

list_sources()                     # All sources
list_sources(category: "frontend") # Filter by category

list_indexed

List libraries that have been indexed for search.

list_indexed()

Built-in Sources

| Name | Category | Description | |------|----------|-------------| | react | frontend | React documentation | | nextjs | frontend | Next.js documentation | | tailwindcss | frontend | Tailwind CSS | | vite | frontend | Vite build tool | | hono | backend | Hono web framework | | fastapi | backend | FastAPI framework | | express | backend | Express.js framework | | anthropic | ai | Anthropic Claude API | | openai | ai | OpenAI API | | langchain | ai | LangChain framework | | supabase | database | Supabase documentation | | drizzle | database | Drizzle ORM | | prisma | database | Prisma ORM |

Custom Sources

Add to ~/.config/docpull-mcp/sources.yaml:

sources:
  my-docs:
    url: "https://docs.example.com"
    description: "My documentation"
    category: "internal"
    maxPages: 200

Manual Ingestion

To ingest docs without the MCP server:

# Ingest all fetched docs
bun run ingest

# Ingest specific library
bun run ingest react

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | DATABASE_URL | For search | PostgreSQL connection string with pgvector | | OPENAI_API_KEY | For search | OpenAI API key for embeddings | | DOCS_DIR | No | Custom docs directory (default: ~/.local/share/docpull-mcp/docs) |

Cost

• Fetching: Free (uses docpull)
• Indexing: ~$0.10 per 50,000 words (OpenAI embeddings)
• Searching: ~$0.0001 per query

License

MIT