Documentation Search MCP Server

A high-performance Model Context Protocol (MCP) server built with Python, designed to fetch and convert official library documentation into clean, LLM-friendly Markdown.

It leverages DuckDuckGo for precise searching and selectolax for ultra-fast HTML parsing.

🚀 Features

• Parallel Scraping: Uses asyncio and httpx to fetch multiple documentation pages simultaneously.
• Smart Parsing: Strips headers, footers, and sidebars to provide only the relevant technical content.
• Markdown Conversion: Formats HTML into structured Markdown (headings, code blocks, lists).
• Dynamic Registry: Support for pre-configured libraries (LangChain, OpenAI, etc.) with the ability to dynamically register new sources at runtime.
• Auto-Discovery: Can attempt to find a library's official documentation URL automatically if it's not in the registry.

🛠️ Installation

1. Clone the repository:

   git clone <your-repo-url>
   cd mcp_documentation
   

2. Create a virtual environment and install dependencies:

   Using python and pip:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate
   pip install -r requirements.txt
   

   Or, using uv as a faster drop-in replacement:

   # Create a virtual environment (optional if you prefer uv-managed envs)
   uv venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate
   
   # Install from requirements.txt using uv's pip interface
   uv pip install -r requirements.txt
   

⚙️ Configuration

Adding Extra Documentation via Environment

You can add custom documentation sources without modifying the code by setting the MCP_EXTRA_DOCS environment variable:

export MCP_EXTRA_DOCS='{"fastapi": "https://fastapi.tiangolo.com", "flask": "https://flask.palletsprojects.com"}'

### Claude Desktop Integration
Add this to your `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "doc-search": {
      "command": "/path/to/your/venv/bin/python",
      "args": ["/path/to/mcp_documentation/main.py"],
      "env": {
        "MCP_EXTRA_DOCS": "{\"fastapi\": \"https://fastapi.tiangolo.com\"}"
      }
    }
  }
}

🛠️ Tools

search_docs(library, query)

Searches the specified library for the query.

• Example: search_docs(library="langchain", query="how to use multi-agent systems")
• Behavior: If the library isn't registered, it will attempt to find the documentation URL via DuckDuckGo and auto-register it for the current session.

register_new_library(name, base_url)

Manually add a new documentation site to the server's knowledge during a session.

• Example: register_new_library(name="pytorch", base_url="https://pytorch.org/docs")

list_supported_libraries()

Returns a list of all libraries currently in the registry.

🏗️ Project Structure

• main.py: Entry point for the server.
• src/server.py: FastMCP server definition and tool implementation.
• src/parser.py: Logic for cleaning HTML and converting it to Markdown.
• src/search.py: DuckDuckGo search integration.
• src/config.py: Registry management and default configurations.

📄 License

MIT