tuannvm/codex-mcp-server — reverse-engineered prompt

Reverse engineered prompt

Build me a small TypeScript MCP server that lets Claude Code talk to the OpenAI Codex CLI inside the editor. I want it to feel simple to install and use, so after someone installs Codex CLI and logs in with their OpenAI API key, they can add this server and start asking Codex to explain code, refactor code, review local changes or PR changes, and do web searches without leaving Claude Code or Cursor.

Please include a main codex tool for analysis and generation, a review tool for uncommitted changes, branches, or commits, a websearch tool, plus simple utilities like ping, help, and a way to list active sessions. Session support matters, so users should be able to continue the same conversation by reusing a session id. It should support options like model choice, reasoning effort, sandbox mode, callback URI, and return structured output metadata like thread id when available.

Make it production ready, tested, and easy to run with npx. If anything is unclear, check current MCP and Codex CLI docs online.

Want more depth? Deep Reverse