I built claude-anyteam because I had impulsively subscribed to both OpenAI's and Anthropic's pro and max subs and, since I mainly work within CC, especially its Agent Teams, my codex quota mostly remained untouched. I wondered if I could get Codex, and its harness, to register and participate as a native Agent within CC, and how far I could take it.

There were pre-existing solutions out there, but they all focused on having an API proxy sit between where, in this case, codex would pretend to be Claude.

But I wanted codex to join as an actual team member that spoke CC's protocol directly and supported things like inbox polling, task claiming, completion messages, idle/shutdown lifecycle, I/O inbox, mid-task messaging, etc.

In claude-anyteam, Codex teammates show up in the Agent Teams UI and can work inside the same work, indistinguishible from CC teammates. I was initially worried that it being a different model meant that it would use the Agent Teams protocol less effectively, but I ended up surprised with how native it felt.

Peers can inject messages into a running Codex task, through the Codex App Server mode, and each new task can fork from the previous Codex thread to the teammate keeps context across task.

Today there's only support for Codex, but this has worked so well in my personal setup that I'm planning on adding support for more agents/models.

Happy to learn what people think about this idea/approach! And if anyone has had this same problem and how they've solved it.