DOC-Intelligence-MCP-SERVER

A Model Context Protocol (MCP) server that enables LLMs (like Claude) to dynamically search, scrape, and query libraries' official documentation using Google Serper and Groq.

Features

• Google Serper API Integration: Searches documentation sites for up-to-date information.

• Fast Web Scraping: Extracts page content cleanly using trafilatura.

• LLM Content Cleaning: Automatically cleans and strips HTML bloat from documentation pages using Groq (llama-3.1-8b-instant).

• Claude Integration: Instantly works as a tool within Claude Desktop or any other MCP client.

Related MCP server: Library Docs MCP Server

Supported Libraries (Configured Domains)

• uv (docs.astral.sh/uv)

• openai (platform.openai.com/docs)

• langchain (python.langchain.com/docs)

• llama-index (docs.llamaindex.ai/en/stable)

Setup

1. Requirements

Ensure you have uv installed.

2. Configure Environment Variables

Copy .env.example to .env and fill in your API keys:

cp .env.example .env

Inside .env:

• SERPER_API_KEY: Get one from Serper.dev

• GROQ_API_KEY: Get one from Groq Console

3. Run Locally (Testing)

You can run the test client to verify operations:

uv run client.py

Claude Desktop Integration

To add this tool to your Claude Desktop client, edit your configuration file:

• Windows Path: %APPDATA%\Claude\claude_desktop_config.json

• macOS Path: ~/Library/Application Support/Claude/claude_desktop_config.json

Choose one of the integration methods below:

Option 1: Run directly from GitHub (Quickest, no cloning required)

This runs the MCP server directly from GitHub. Add this JSON snippet under "mcpServers":

{
  "mcpServers": {
    "docs-intelligence": {
      "command": "uv",
      "args": [
        "run",
        "https://raw.githubusercontent.com/SARAN-KUMAR-S/DOC-Intelligence-MCP-SERVER/main/mcp_server.py"
      ],
      "env": {
        "SERPER_API_KEY": "YOUR_SERPER_API_KEY",
        "GROQ_API_KEY": "YOUR_GROQ_API_KEY"
      }
    }
  }
}

Option 2: Clone and Run Locally (Best for making code changes)

If you cloned the repository locally, create a .env file containing your API keys in the root directory, and configure the path under "mcpServers":

{
  "mcpServers": {
    "docs-intelligence": {
      "command": "uv",
      "args": [
        "--directory",
        "ABSOLUTE_PATH_TO_CLONED_FOLDER",
        "run",
        "mcp_server.py"
      ]
    }
  }
}

Restart Claude Desktop, and the get_docs tool will be ready to use!