🤖 MCP GitHub PR Reviewer

What is MCP GitHub PR Reviewer?

MCP GitHub PR Reviewer is a Node.js server that implements part of the Model Context Protocol (MCP) to automate code review for GitHub Pull Requests. It fetches PR details from GitHub, sends them to a Large Language Model (LLM) via OpenRouter.ai or directly to Anthropic/OpenAI, and saves the review results as both HTML and Markdown files.

How to use MCP GitHub PR Reviewer?

Clone the repository, install dependencies with npm install, configure environment variables in a .env file (including GITHUB_TOKEN and one of OPENROUTER_API_KEY or ANTHROPIC_API_KEY), and start the server with npm start. Then send a POST request to http://localhost:3000/review-pull-request with a JSON body containing the prLink field.

Key features of MCP GitHub PR Reviewer

• Automatic code review via LLMs for bugs, security, and best practices
• Integration with GitHub API to fetch PR diffs and descriptions
• Flexible LLM support through OpenRouter.ai or direct Anthropic/OpenAI APIs
• Informative output in both HTML and Markdown formats with public links
• Modular server backend ready for integration with clients like bots or web apps
• Configurable AI model and prompt customization

Use cases of MCP GitHub PR Reviewer

• Quickly obtain initial feedback on new pull requests without manual review
• Automate code quality checks and flag potential issues early in development
• Provide consistent review summaries for team discussion and logging
• Integrate with CI/CD pipelines or chat bots to notify developers of AI review results

FAQ from MCP GitHub PR Reviewer

What dependencies and runtime are required?

Node.js version 18.x or higher, Git, a GitHub Personal Access Token, and an API key for either OpenRouter.ai, Anthropic, or OpenAI. The server runs locally and exposes a single POST endpoint.

How does the server handle errors with GitHub or the AI API?

Common errors include 404 (PR not found), 403 (access denied), and 500 (AI API failure). Verify that the PR link is valid, the GitHub token has correct repo scope, and that the AI API key and model are configured correctly with sufficient balance.

Where are the review files stored?

Both HTML and Markdown review files are saved locally in a reviews/ directory at the project root. The server also provides a public HTML URL for each review.

How can I customize the AI model or the review prompt?

Model selection is set via the OPENROUTER_MODEL or ANTHROPIC_MODEL environment variable. The system and user prompts used for review can be modified in the getAIReview function inside server.js.

What transport and authentication does the server use?

The server uses HTTP transport (POST requests) and expects a JSON body. It authenticates to GitHub via a Personal Access Token and to the AI provider via API keys stored in environment variables. No authentication is required for the local server endpoint itself.