Hi Tibo / Codex team — flagging a third-party project that may be worth reviewing:
miuuyy/codex-chatgpt-web
The project appears to use ChatGPT Web as a model backend for Codex, allowing users who have exhausted their Codex usage limits to continue running Codex tasks against their remaining ChatGPT Web quota.
The newer versions go further by reconnecting the Web model to the Codex harness through MCP, including filesystem, shell, images, approvals and other tool capabilities.
I don’t think the core issue is that this is an unofficial client, nor do I think local MCP itself should be restricted because of it.
The concern is the usage-accounting boundary:
Codex limit reached → route the workload through ChatGPT Web → reconnect it to the Codex harness → continue essentially the same Codex workload outside the Codex usage pool.
Even if the project does not technically modify a rate limit, this seems capable of turning separate product quotas into substitute paths for one another.
It is also now being promoted publicly as a way to get “unlimited free Codex,” which could drive substantially more abuse than the original project intended.
Might be worth having the Codex / abuse team take a look before this pattern becomes widespread.
I’d much rather see the Web-to-Codex bypass addressed specifically than have legitimate local MCP users lose functionality because other people abused it.
Repo:
github.com/miuuyy/codex-chat…