I built Mac Developer Bridge because I kept ending up as the human copy/paste layer between a strong ChatGPT conversation and the Mac where the real work lived. The workflow I wanted was simple: “Find the Codex session I was working on yesterday, recover the context, inspect the live repo, fix what’s broken, and push the result.” I wanted that to happen from normal ChatGPT without starting another model loop or relaying files and terminal output by hand. So the bridge exposes the Mac over MCP: shell execution, unrestricted filesystem operations, real PTY sessions, detached jobs with logs, and read-only access to persisted Codex threads. The bridge itself makes no model calls. ChatGPT does the reasoning; the Mac does the work. One warning matters more than the feature list: this is intentionally not sandboxed. It runs with the effective permissions of your logged-in macOS user. I added an explicit unlock latch, local audit log, a kill switch, and detailed security documentation, but this is for people who deliberately want that permission model. It’s free, MIT licensed, and the core bridge has zero npm runtime dependencies. I’d love feedback on two things: where the setup still feels too fiddly, and what local workflow you’d want normal ChatGPT to take over first.

Mac Developer Bridge

Give ChatGPT a real terminal on your Mac

Give ChatGPT a real terminal on your Mac