git-mcp

Purpose

GitMCP is designed to solve the common "code hallucination" problem in AI programming tools when generating code. It is a free, open-source remote MCP (Model Context Protocol) server that can turn any public GitHub project (including repositories and GitHub Pages sites) into a documentation hub accessible to AI assistants, enabling AI to obtain the latest project documentation and source code, even if the model has never encountered the project before.

Main Capabilities

• Retrieve documentation and code for any GitHub project: Provide AI assistants with the latest documentation and code examples for a project.
• Intelligent search: Built-in intelligent search capability can precisely find the required information in documentation and code, avoiding excessive token consumption.
• Reduce hallucinations: Generate answers based on real documentation and code, improving accuracy and code correctness.
• Embedded chat: Can chat directly with repository documentation in the browser for a quick start.
• Zero configuration, cloud-based operation: No download, installation, or registration required; simply add the GitMCP URL as an MCP server in your IDE.
• Open, free, privacy-preserving: Open source and completely free to use, does not collect personal information, does not store queries, and supports self-hosting.

How to Use

Step 1: Choose the server type

• For a specific GitHub repository:gitmcp.io/{owner}/{repo}
• For a GitHub Pages site:{owner}.gitmcp.io/{repo}
• Universal dynamic endpoint (can access any repository):gitmcp.io/docs

Replace {owner} with the GitHub username or organization name, and{repo} with the repository name. You can also use the conversion tool on the landing page to convert a GitHub URL into an MCP URL.

Step 2: Connect the AI assistant

For example, with Cursor, configure it in ~/.cursor/mcp.json ; for Claude Desktop, go to Settings > Developer > Edit Config and use the following configuration:

{
  "mcpServers": {
    "gitmcp": {
      "command": "npx",
      "args": ["mcp-remote", "https://gitmcp.io/{owner}/{repo}"]
    }
  }
}

Other tools (Windsurf, VSCode, Cline, Highlight AI, Augment Code, Msty Studio, etc.) also provide corresponding configuration instructions.

How It Works

1. The user provides the GitMCP URL to the AI assistant; 2. The user asks a question related to documentation or code; 3. With the user's approval, the AI sends a request to GitMCP; 4. GitMCP executes the request and returns data; 5. The AI generates a more accurate answer based on real data.

Available Tools

• fetch_<repo-name>_documentation: Retrieve the repository's main documentation (such as llms.txt).
• search_<repo-name>_documentation: Perform intelligent search within the documentation.
• fetch_url_content: Retrieve the content of external links referenced in the documentation and convert it into an easy-to-read format.
• search_<repo-name>_code: Use GitHub code search to find code examples or implementation details in the repository. When using the dynamic endpoint (gitmcp.io/docs), the tool names are fetch_generic_documentation、search_generic_code、search_generic_documentation, and the target repository must also be specified.

Use Cases

• When using niche, newly released, or rapidly changing libraries, obtain the latest API usage and reliable code examples.
• When you need AI to answer questions based on specific repository documentation.
• Quickly understand project documentation through embedded chat.

Limitations and Notes

• GitMCP only accesses content that is already public, and only accesses it when the user queries, and does not automatically crawl repositories.
• Before accessing GitHub Pages sites, it checks robots.txt and follows the site owner's instructions.
• Does not permanently store project or content data.
• Currently, the documentation priority is: AI-optimized project documentation, README.md.
• This project is provided "as is" without any warranty; it is not affiliated with GitHub or the AI tools mentioned.
• Licensed under the Apache License 2.0.