Namita-Singroha/mcp-gitreview

MCP GitReview Server

🚀 A lightweight Model Context Protocol (MCP) server for exploring GitHub repositories over stdio.

This project enables AI agents (e.g., Claude Desktop) to interact with GitHub repos in a structured way — listing commits, branches, PRs, and repo metadata — without exposing raw git or GitHub API complexity.

✨ Features

• Repository selection → Set or switch repositories at runtime
• Authentication → Add your GitHub Personal Access Token (PAT) for private repos
• GitHub data tools:
  • View last 10 commits
  • List all branches
  • Get the latest commit message
  • Fetch repo metadata (stars ⭐, forks 🍴, open issues 🐞)
  • List open pull requests

All these tools are exposed as MCP endpoints that an LLM can call directly.

📂 Project Structure

internal/
  github/   → GitHub API client
  server/   → Context handling (repo + token)
  tools/    → MCP tool definitions
main.go     → Entry point (starts the MCP server)

🔧 Usage

1. Build the server

go build -o mcp-gitreview .

2. Run via MCP

Example config (Claude Desktop / MCP client):

{
  "mcpServers": {
    "gitreview": {
      "command": "/path/to/mcp-gitreview"
    }
  }
}

👉 No default repo is required. If a tool is called without a repo set, the server will politely ask the LLM to set one using set_repo.

📸 Demo: Claude Desktop interacting with the MCP server by invoking MCP Tools. The user asks in natural language, and the server communicates back in simple English, describing repository features clearly.

🤖 Example Flow

1. Ask: “List the branches in modelcontextprotocol/go-sdk”
2. LLM realizes no repo is set → calls set_repo
3. Server stores it → responds: “Repo set to modelcontextprotocol/go-sdk”
4. LLM calls list_branches → gets actual branch names.