codex-mcp

OpenAI Codex CLI as an MCP server. Use GPT-5.4 to write code from Claude Code, Cursor, Windsurf, or any MCP client.

> Built this before the official plugin existed. This does more.

Exposes 7 tools that let any MCP client dispatch tasks to the Codex CLI (GPT-5.4):

| Tool | Description | |------|-------------| | codex_build | Write code full-auto in any directory | | codex_review | Code review a repository | | codex_query | Ask questions with read-only codebase access | | codex_resume | Resume a previous Codex session | | codex_sessions | List recent sessions | | codex_skills | Browse reusable task templates | | codex_memories | View project-specific memories |

• Works in any MCP client — Claude Code, Cursor, Windsurf, anything with MCP support
• Multi-repo — specify any cwd, not just the current project
• Session management — resume interrupted builds
• Model selection — gpt-5.4, o3, o4-mini per task
• Skills + memories — expose Codex's knowledge base as tools

npm install -g @openai/codex
codex login  # authenticate with OpenAI

pip install codex-mcp

Add to ~/.claude/claude_desktop_config.json or .mcp.json in your project:

{
  "mcpServers": {
    "codex": {
      "command": "python3",
      "args": ["-m", "codex_mcp"]
    }
  }
}

Or point directly at the script:

{
  "mcpServers": {
    "codex": {
      "command": "python3",
      "args": ["/path/to/codex-mcp/server.py"]
    }
  }
}

Same config in .cursor/mcp.json or .windsurf/mcp.json.

Once installed, Claude (or any MCP client) can:

Use codex_build to implement a login page in /root/myapp

Use codex_review to review /root/myapp for security issues

Use codex_query to explain the auth flow in /root/myapp

| Env var |