Codex Review MCP Server

Get AI-powered code reviews without leaving your conversation. Keep coding with your AI assistant, and when you need a second opinion, just ask for a review - all in the same context window.

Why This Exists

You're coding with an AI assistant. It makes changes. You want a quick review. Normally you'd copy/paste code into another chat, wait, copy feedback back. That breaks your flow.

With this MCP server, your AI assistant can call out for a code review right in the middle of your conversation. No context switching. No copy/paste. The review comes back in the same chat.

Features

• 🎯 Zero Configuration - Just add your OpenAI API key
• 📋 Respects Your Cursor Rules - Automatically reads .cursor/rules/ and enforces YOUR coding standards
• 🔍 Automatically reviews your uncommitted changes
• 🤖 Powered by GPT-5 Codex for intelligent code analysis
• 📊 Returns actionable feedback in Markdown format
• ⚡ Works with any git repository
• 🔄 Stays in your conversation context

Installation

For Cursor Users

Add the following configuration to your Cursor MCP settings (~/.cursor/mcp.json or %USERPROFILE%\.cursor\mcp.json on Windows):

{
  "mcpServers": {
    "codex-reviewer": {
      "command": "npx",
      "args": ["-y", "codex-review-mcp@latest"],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key-here",
        "CODEX_MODEL": "gpt-5-codex",
        "WORKSPACE_ROOT": "${workspaceFolder}"
      }
    }
  }
}

Important: Replace your-openai-api-key-here with your actual OpenAI API key.

For Claude Desktop Users

Add to your Claude Desktop MCP configuration:

{
  "mcpServers": {
    "codex-reviewer": {
      "command": "npx",
      "args": ["-y", "codex-review-mcp@latest"],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key-here",
        "CODEX_MODEL": "gpt-5-codex"
      }
    }
  }
}

Usage

Once configured, just ask your AI assistant:

"Do a code review with codex-reviewer"

That's it! The MCP server automatically figures out what to review:

• If you have uncommitted changes: Reviews your working tree vs HEAD (what you just coded)
• If your working tree is clean: Reviews your current branch vs the default branch (main/master)
• If you're on the default branch with no changes: Reports "No changes to review"

Example conversation:

You: "Add a login form with email validation"
AI: [makes changes]
AI: "Let me get a code review on this..."
AI: [calls codex-reviewer MCP - automatically reviews uncommitted changes]
AI: "The review found a potential issue with the email regex..."

No need to specify what to review - it just works!

Advanced Options (Optional)

Most users won't need these, but they're available if you want more control:

• target (default: "head"): What to review
  • "head": All uncommitted changes (default - what you just coded)
  • "staged": Only staged changes (git added files)
  • "range": Specific commit range (requires baseRef and headRef)
• focus: Ask the reviewer to focus on something specific (e.g., "security issues")
• paths: Review only specific files
• baseRef / headRef: For reviewing commit ranges
• maxTokens: Limit the length of the review
• workspaceDir: Override auto-detection of repository location

How It Works

1. Detects Repository: Automatically finds your git repository root
2. Collects Diff: Runs git diff to get the changes
3. Gathers Context: Reads your .cursor/rules/, CODE_REVIEW.md, .eslintrc, and other project files
4. AI Review: Sends to GPT-5 Codex with YOUR project's coding standards
5. Returns Feedback: Provides actionable Markdown feedback that respects YOUR rules

Requirements

• Node.js 18+ (for running the MCP server)
• Git repository
• OpenAI API key with access to GPT-5 Codex
• Cursor or Claude Desktop (or any MCP-compatible client)

Environment Variables

• OPENAI_API_KEY (required): Your OpenAI API key
• CODEX_MODEL (optional, default: "gpt-5-codex"): The model to use for reviews
• WORKSPACE_ROOT (optional): Path to your workspace (auto-detected in most cases)

Development

Local Development Setup

# Clone the repository
git clone <your-repo-url>
cd codex-review-mcp

# Install dependencies
npm install

# Build
npm run build

# For local testing, update your MCP config to point to the local build:
{
  "codex-reviewer": {
    "command": "node",
    "args": ["/absolute/path/to/codex-review-mcp/dist/mcp-server.js"],
    "env": {
      "OPENAI_API_KEY": "your-key",
      "CODEX_MODEL": "gpt-5-codex",
      "WORKSPACE_ROOT": "${workspaceFolder}"
    }
  }
}

License

MIT

Contributing

Contributions are welcome! Please open an issue or PR.