Git MCP

Overview

Git MCP offers the simplest possible way to give AI agents access to real GitHub repository code. No setup, no tokens, no installation — just prepend “gitm.cp/” to any GitHub URL and you have an MCP endpoint that serves the actual repository contents.

How It Works

1. Find a repo URL — e.g., github.com/vercel/next.js
2. Prepend gitm.cp/ — becomes gitm.cp/github.com/vercel/next.js
3. Add as MCP server — Use the URL as a remote MCP endpoint
4. Query real code — Agent reads actual repo files

Use Cases

• Library exploration — AI reads actual source to understand APIs
• Code examples — Find real usage examples from the library
• Bug reporting — Agent references actual code when filing issues
• Learning — Understand how a library works by reading its source

Getting Started

{
  "mcpServers": {
    "nextjs": {
      "command": "npx",
      "args": ["mcp-remote", "https://gitm.cp/github.com/vercel/next.js"]
    }
  }
}

Example

User: "How does Next.js implement loading.tsx?"

Agent → Git MCP:
  - search_files("loading", path: "packages/next/src")
  - read_file("packages/next/src/server/app-render/create-component-tree.tsx")

Agent explains the actual implementation from source code.

Alternatives

• GitHub MCP Server — Full GitHub API access (PRs, issues)
• Context7 — Curated documentation (not source code)
• Serena — Semantic code navigation for local projects