GitHub MCP Server

A custom Model Context Protocol (MCP) server that connects Claude Desktop directly to GitHub. Claude autonomously decides which tools to call based on a plain English goal — no manual orchestration needed.

---

What it does

Instead of you telling Claude how to fetch data, you give Claude a goal and it figures out which tools to call, in what order, and how many times.

Example:

You type: "Review all open PRs in expressjs/express and flag any that touch auth code"

Claude autonomously:

1. Calls list_open_prs → gets 10 PRs
2. Calls get_pr_diff on each one
3. Calls get_pr_metadata to check which files changed
4. Calls post_pr_comment on the ones that touch auth

You didn't write a single line of orchestration code.

---

Tools exposed

Tool	Description
list_open_prs	List all open pull requests in a repository
get_pr_diff	Fetch the full code diff for a specific PR
get_pr_metadata	Get title, author, description, files changed
post_pr_comment	Post a review comment on a PR

---

Tech stack

Layer	Technology
Protocol	Model Context Protocol (MCP)
Runtime	Python 3.12
GitHub integration	GitHub REST API v3
Transport	stdio (standard MCP transport)
Client	Claude Desktop

---

CCA exam domains covered

Domain	Weight	How this project covers it
Tool Design & MCP Integration	18%	Full MCP server with 4 production-ready tools
Agentic Architecture & Orchestration	27%	Claude autonomously orchestrates multi-step GitHub workflows
Claude Code Configuration & Workflows	20%	MCP server config, stdio transport, tool schema design

---

Project structure

github-mcp-server/
├── server.py      # MCP server with all 4 tools
├── .env.example   # Environment variable template
└── requirements.txt

---

Setup

1. Requirements

• Python 3.10+ (3.12 recommended)
• Claude Desktop installed (claude.ai/download)
• GitHub personal access token

2. Clone and install

git clone https://github.com/Medhavi1101/github-mcp-server
cd github-mcp-server
python3.12 -m venv venv
source venv/bin/activate
pip install "mcp[cli]" httpx python-dotenv

3. Configure environment

cp .env.example .env

Edit .env:

GITHUB_TOKEN=ghp_your-token-here

4. Connect to Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "github": {
      "command": "/path/to/your/venv/bin/python3",
      "args": ["/path/to/github-mcp-server/server.py"],
      "env": {
        "GITHUB_TOKEN": "ghp_your-token-here"
      }
    }
  }
}

Restart Claude Desktop. You'll see the tools available in the chat.

5. Test it

In Claude Desktop, type:

List open PRs in expressjs/express

Claude will call your list_open_prs tool automatically and return live results.

---

How MCP differs from a regular API

Regular API	MCP Server
You call the API	Claude calls the tools
You write orchestration logic	Claude decides the order
Fixed workflow	Dynamic, goal-driven
You handle errors and retries	Claude adapts autonomously

---

Related project

This MCP server pairs with the AI Code Reviewer API — a FastAPI backend that uses Claude tool use to return structured code reviews. Together they form a complete agentic code review system.

---

• LinkedIn