Codex MCP Server
An MCP server wrapper that bridges Claude (and other MCP clients) with OpenAI's Codex CLI, letting you harness GPT-powered code generation, analysis, refactoring, and execution directly from your preferred AI coding environment.
Overview
Codex MCP Server is a lightweight, open-source Model Context Protocol (MCP) server that wraps OpenAI's official Codex CLI. It acts as a bridge, allowing MCP clients such as Claude Desktop/Code, Cursor, Windsurf, and others to call Codex as a tool for advanced coding tasks.
This enables hybrid AI workflows where Claude handles high-level reasoning and strategy while delegating heavy code generation, refactoring, analysis, and execution to Codex's powerful GPT-based engine.
Popular community project maintained by tuannvm and others.
Key Features
- Full integration with OpenAI Codex CLI capabilities
- Exposes Codex commands (explain, refactor, generate, analyze, exec, etc.) as structured MCP tools
- Supports
codex execfor safe, sandboxed code execution and edits - Streams progress updates and structured patch outputs
- Easy one-command installation via
npx - Works with both local STDIO and configurable environments
- Cross-platform (macOS, Linux, Windows)
- Secure approval flows for file modifications
Installation & Setup
Quick Start (for Claude Code)
claude mcp add codex-cli -- npx -y codex-mcp-server
Alternative repositories / forks
- https://github.com/tuannvm/codex-mcp-server (primary)
- https://github.com/cexll/codex-mcp-server (enhanced version)
- Built-in Codex MCP mode:
codex mcp-server(native support in recent Codex CLI)
After adding, restart your MCP client and confirm the server appears in the tools list.
Use Cases
- Hybrid Agentic Coding: Use Claude for planning and Codex for high-quality code output
- Code Review & Refactoring: Ask Claude to review a codebase, then delegate complex refactors to Codex
- Debugging Stuck Tasks: When one model gets stuck, hand off to the other via MCP
- Large-scale Analysis: Leverage Codex's context window and reasoning on massive projects
- Multi-model Orchestration: Combine strengths of Anthropic and OpenAI models in a single workflow
- CI/CD Automation and batch code transformations
Requirements
- OpenAI Codex CLI installed (
npm i -g @openai/codexor Homebrew) - Valid OpenAI API access (Codex uses GPT models)
- An MCP-compatible client (Claude Code/Desktop recommended)
Resources
- GitHub: https://github.com/tuannvm/codex-mcp-server
- Official Codex MCP Documentation: https://developers.openai.com/codex/mcp/
- Native Codex as MCP Server: Run
codex mcp-serverdirectly
Codex MCP Server is one of the most practical tools for developers wanting to combine the best of Claude's reasoning with Codex's coding prowess in the growing MCP ecosystem.