GitHub PR Analysis MCP Server

GitHub PR Analysis MCP Server

This project implements a Model Context Protocol (MCP) server that analyzes GitHub Pull Requests and optionally creates a structured Notion page with the analysis results. It is designed to be used locally with Claude CLI using the stdio MCP transport.

🚀 Features: Analyze GitHub Pull Requests using GitHub API Extract PR metadata, commits, files, and code changes Generate structured AI-based PR analysis Optionally create a Notion page with the analysis Runs locally via MCP stdio Integrates seamlessly with Claude CLI

🧠 Architecture Overview: Claude CLI │ │ (MCP stdio) ▼ MCP Server (Python) ├── GitHub API (PR data) ├── AI Analysis Logic └── Notion API (Page creation)

📁 Project Structure: MCP-STDIO/ ├── pr_analyzer.py # MCP server entry point ├── github_integration.py # GitHub PR fetching logic ├── requirements.txt # Python dependencies └── .venv/ # Virtual environment

🔑 Required Environment Variables:

The application relies on the following environment variables:

Variable Description GITHUB_TOKEN GitHub Personal Access Token NOTION_API_KEY Notion integration secret NOTION_PAGE_ID Parent Notion page ID

🔐 How to Create GITHUB_TOKEN: Go to 👉 https://github.com/settings/tokens

Click Generate new token (classic) Select scopes: ✅ repo ✅ read:user Generate token and copy it

🧾 How to Create NOTION_API_KEY: Go to 👉 https://www.notion.so/my-integrations

Click New integration Name it (e.g. PR Analyzer) Select your workspace Copy the Internal Integration Secret

📄 How to Create NOTION_PAGE_ID: Create a page in Notion (this will be the parent page) Share the page with your integration: Click Share Invite your integration Copy the page URL:

https://www.notion.so/AI-PR_ANALYSIS-2*************************

Extract the page ID (last 32 characters): 2***********************

📦 Dependencies (requirements.txt): Package Purpose requests Communicates with GitHub & Notion REST APIs python-dotenv Loads environment variables from .env fastmcp MCP server framework

🛠 Setup Instructions: 1️⃣ Create Virtual Environment cd MCP-STDIO python -m venv .venv source .venv/bin/activate

2️⃣ Install Dependencies pip install -r requirements.txt

3️⃣ Verify Server Runs Manually .venv/bin/python pr_analyzer.py

✅ This should start the MCP server without errors.

🤖 Using the MCP Server with Claude CLI 1️⃣ Ensure Claude CLI is Installed claude --version

Expected output:

2.x.x (Claude Code)

2️⃣ Configure MCP Server (~/.claude.json) { "mcpServers": { "github_pr_analysis": { "type": "stdio", "command": "full path of your .venv/bin/python ", "args": ["pr_analyzer.py"], "cwd": "full path of your pr_analyzer.py" } } }

⚠️ Make sure: command points to the virtualenv python cwd is the folder containing pr_analyzer.py

3️⃣ Restart Claude CLI claude 4️⃣ Verify MCP Server is Connected Inside Claude CLI: /mcp You should see: github_pr_analysis · ✔ connected 🧪 Example Usage in Claude Analyze this PR: https://github.com/org/repo/pull/123

Claude will: Fetch PR details Perform analysis

Ask: “Would you like me to create a Notion page for this analysis?”

Reply with: yes → creates Notion page no → skips creation

🐞 Debugging Tips Add logs in MCP server: print("Debug message", file=sys.stderr)

Check Claude MCP logs: claude --debug

Verify paths: pwd ls pr_analyzer.py

✅ Summary MCP server runs locally via stdio Claude CLI acts as the client GitHub PRs are analyzed automatically Notion pages are created on user confirmation Secure via environment variables

children=[{ "object": "block", "type": "paragraph", "paragraph": { "rich_text": [{ "type": "text", "text": {"content": content} }] } }]