Feature Request: Official Codex MCP Server — make Codex universal coding infrastructure

Hey everyone,

Wanted to propose something that I think is a big commercial opportunity for OpenAI and a massive win for the dev community.

**The idea:** Ship an official Codex MCP (Model Context Protocol) server.

MCP is an open protocol and OpenAI already supports it in the Agents SDK. But Codex itself doesn’t have an MCP server which means developers who want to use Codex as a coding backend in multi-model workflows have to wire it up manually via raw API calls every time.

**The bigger picture — Codex as infrastructure:**

Think of it like Stripe. Stripe doesn’t care which app is processing the payment, just that the API gets called. Codex could sit in that same position: the best-in-class coding execution layer, pluggable into any agent stack regardless of who’s orchestrating.

Right now MCP-compatible frontends include Claude.ai, ChatGPT, Cursor, Windsurf, and more. An official Codex MCP server means every one of those becomes a potential Codex API customer — without OpenAI having to compete on the orchestration layer at all.

**Why devs want this:**

Many of us use different AI frontends for different tasks. Having Codex available as a drop-in MCP tool means we can route coding tasks to it from whatever environment we’re already in. Right now that friction is enough to make us reach for alternatives.

Would love to know if this is on the roadmap. Drop a reply if you’d want this too :backhand_index_pointing_down:

Welcome to the dev Community! @Ayushman2010

This is a pretty compelling idea honestly.

I can see why people would want an official Codex MCP server instead of rebuilding the same API wiring over and over themselves. The “Codex as infrastructure” angle also makes sense, especially now that MCP support is showing up across more tools and clients.

Being able to call Codex from whatever frontend or orchestration layer someone already uses would definitely reduce friction for multi-model workflows.

I’ll pass this along internally as product feedback since the use case and reasoning here are both pretty clear.

~ Smith