Code Review MCP Server

English | 中文

MCP (Model Context Protocol) server for code review. Enables AI assistants to review GitHub/GitLab Pull Requests and Merge Requests.

Keywords: MCP server, AI code review, GitHub pull request review, GitLab merge request, Cursor IDE, Claude Desktop, Model Context Protocol, automated PR comments.

✨ Features

• 🔍 Multi-platform: Supports both GitHub and GitLab (including self-hosted)
• 🚀 Multiple Transports: Supports stdio, SSE, and WebSocket protocols
• 📦 Easy Install: Quick install via uvx or pip
• 🐳 Containerized: Docker image available
• ☁️ Cloud Deploy: One-click Smithery deployment
• 🔒 Security First: Environment variable configuration, no data persistence

🚀 Quick Start

Option 1: Using uvx (Recommended)

# Run directly, no installation needed
uvx code-review-mcp

Option 2: Using pip

pip install code-review-mcp

# Run the server
code-review-mcp

# (Optional) Install Cursor rules to your project
code-review-mcp init-rules

Option 3: From Source

git clone https://github.com/OldJii/code-review-mcp.git
cd code-review-mcp
pip install -e .
code-review-mcp

🔧 Configuration

Environment Variables

Getting Tokens

GitHub

# Option 1: Using gh CLI (Recommended)
brew install gh
gh auth login

# Option 2: Manual Token Creation
# Visit https://github.com/settings/tokens
# Create Personal Access Token with 'repo' scope
export GITHUB_TOKEN="your-token-here"

GitLab

# Option 1: Using glab CLI (Recommended)
brew install glab
glab auth login

# For self-hosted GitLab
glab auth login --hostname gitlab.yourcompany.com

# Option 2: Manual Token Creation
# Visit GitLab -> Settings -> Access Tokens
# Create token with 'api' scope
export GITLAB_TOKEN="your-token-here"
export GITLAB_HOST="gitlab.yourcompany.com"  # For self-hosted

📱 Client Configuration

Cursor

Edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "code-review": {
      "command": "uvx",
      "args": ["code-review-mcp"],
      "env": {
        "GITHUB_TOKEN": "your-github-token",
        "GITLAB_TOKEN": "your-gitlab-token"
      }
    }
  }
}

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "code-review": {
      "command": "uvx",
      "args": ["code-review-mcp"],
      "env": {
        "GITHUB_TOKEN": "your-github-token",
        "GITLAB_TOKEN": "your-gitlab-token"
      }
    }
  }
}

SSE Mode (Remote Deployment)

# Start SSE server
code-review-mcp --transport sse --port 8000

Client configuration:

{
  "mcpServers": {
    "code-review": {
      "url": "http://your-server:8000/sse"
    }
  }
}

WebSocket Mode (Remote Deployment)

# Start WebSocket server
code-review-mcp --transport websocket --port 8000

Client configuration:

{
  "mcpServers": {
    "code-review": {
      "url": "ws://your-server:8000/ws"
    }
  }
}

🐳 Docker Deployment

Build Image

docker build -t code-review-mcp .

Run Container

stdio mode

docker run -i --rm \
  -e GITHUB_TOKEN="your-token" \
  code-review-mcp

SSE mode

docker run -d --rm \
  -e GITHUB_TOKEN="your-token" \
  -p 8000:8000 \
  code-review-mcp --transport sse

🔨 MCP Tools

Rules

Information Retrieval

Adding Comments

💬 Usage Examples

Chat with Cursor or Claude:

Review GitHub PR

Review https://github.com/facebook/react/pull/12345

Review GitLab MR

Review https://gitlab.com/group/project/-/merge_requests/678

Review Self-hosted GitLab MR

Review https://gitlab.yourcompany.com/team/project/-/merge_requests/90

Review Only Specific File Types

Review this PR, only check .py and .js files:
https://github.com/owner/repo/pull/123

🧪 Debugging & Testing

Using MCP Inspector

# Run with MCP Inspector
npx @modelcontextprotocol/inspector uvx code-review-mcp

This launches a web interface where you can:

• View all available tools
• Manually call tools and inspect results
• Debug parameters and responses

Local Development

# Clone repository
git clone https://github.com/OldJii/code-review-mcp.git
cd code-review-mcp

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Debug with Inspector
npx @modelcontextprotocol/inspector python -m code_review_mcp.server

📁 Project Structure

code-review-mcp/
├── src/
│   └── code_review_mcp/
│       ├── __init__.py      # Package entry
│       ├── cli.py           # CLI commands (init-rules, etc.)
│       ├── server.py        # MCP server main logic
│       ├── providers.py     # GitHub/GitLab providers
│       └── rules/           # Bundled Cursor rules
│           ├── code-review.mdc
│           └── code-review-en.mdc
├── pyproject.toml           # Project config & PyPI publishing
├── Dockerfile               # Docker build file
├── smithery.yaml            # Smithery deployment config
├── CHANGELOG.md             # Changelog
├── CONTRIBUTING.md          # Contributing guide
└── README.md                # Documentation

🎯 Cursor Rules (Recommended)

This package includes built-in code review rules for Cursor IDE. Install them to your project with one command:

# Install rules to current project
code-review-mcp init-rules

# Install to a specific directory
code-review-mcp init-rules --target /path/to/project

# Overwrite existing rules
code-review-mcp init-rules --force

# List available rules
code-review-mcp list-rules

After installation, the rules will be available in your project's .cursor/rules/ directory:

• code-review.mdc - Chinese version
• code-review-en.mdc - English version

Custom Project Rules

You can define project-specific review rules that the MCP server loads at runtime. This allows each project to enforce its own coding standards during reviews.

Quick Setup:

# Generate a custom rules template
code-review-mcp init-rules --custom

This creates .code-review-rules/project-rules.md in your project. Edit it with your project-specific conventions, then configure the MCP server to load it:

{
  "mcpServers": {
    "code-review": {
      "command": "uvx",
      "args": ["code-review-mcp"],
      "env": {
        "GITHUB_TOKEN": "your-token",
        "CODE_REVIEW_RULES_DIR": "/absolute/path/to/project/.code-review-rules"
      }
    }
  }
}

How It Works:

• Set CODE_REVIEW_RULES_DIR to a directory containing .md or .mdc files
• Or simply place a .code-review-rules/ directory in your project root (auto-discovered)
• The get_review_rules tool returns both builtin and custom rules
• AI assistants use these rules when performing reviews
• Custom rules supplement (not replace) the builtin review guidelines

Environment Variable:

🤝 Contributing

Contributions welcome! See CONTRIBUTING.md for details.

📄 License

MIT

❓ FAQ

What is code-review-mcp? An MCP server that lets AI assistants (Cursor, Claude, etc.) fetch PR/MR diffs and post inline or general review comments on GitHub and GitLab.

How do I install it? Fastest: uvx code-review-mcp. Or pip install code-review-mcp. See Quick Start.

Does it work with self-hosted GitLab? Yes. Set GITLAB_TOKEN and GITLAB_HOST=gitlab.yourcompany.com.

Which AI clients are supported? Any MCP client — Cursor, Claude Desktop, and custom integrations via stdio, SSE, or WebSocket.

Can I add project-specific review rules? Yes. Run code-review-mcp init-rules --custom or set CODE_REVIEW_RULES_DIR.

Is my code stored on a server? No persistent storage. Tokens are read from environment variables; diffs are fetched on demand from GitHub/GitLab APIs.

Where can AI assistants read a structured summary? See llms.txt in this repository.

🔗 Related Links

• MCP Protocol Documentation
• Smithery Platform
• Cursor Editor
• Claude Desktop