Codex App triggers Xcode 26.3 agent access prompt multiple times
Resolved 💬 7 comments Opened Feb 6, 2026 by dvdchr Closed Mar 29, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
I'm trying out the latest Xcode 26.3 RC with Codex app.
After setting up xcode MCP, the Codex app seems to trigger the "Allow Codex to access Xcode?" prompt multiple times, often causing Codex to time out and leading it to believe the MCP is not set up correctly.
This is not the case with the Codex CLI, where the prompt is only shown once and remains that way for the entire session.
<img width="361" height="444" alt="Image" src="https://github.com/user-attachments/assets/6ca81e22-6609-40a4-b26c-0fcd09e7aef0" />
thread ID 019c30e2-9401-77a1-a6ef-b36df52576e0
7 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
You gotta pump those numbers up. Those are rookie numbers.
<img width="343" height="1076" alt="Image" src="https://github.com/user-attachments/assets/033c1496-ffc2-4b1f-bfb4-cf4f8549d5d5" />
Codex itself told that was a lack of entitlement in the codex macos app.
com.apple.security.automation.apple-eventsCodex :
This is so bad, it makes the codex app unusable with Xcode MCP. Literally have to approve once every 30 s
no fix in sight?
I think this is a duplicate of #10741, and I talked about a temporary workaround there.
Hey! I ran into a similar pattern in our bug knowledge base and thought this might help.
What's happening: Codex App spawns a new MCP server subprocess (or reconnects) for each tool invocation, so Xcode 26.3's per-connection agent authorization prompt fires every time. The CLI reuses a single long-lived MCP connection, so the prompt appears only once. This is NOT a missing entitlement issue — com.apple.security.automation.apple-events controls Apple Events, not Xcode's new MCP agent-access authorization gate.
What worked for us:
The proposed fix correctly identifies the need for a single persistent MCP connection and a retry mechanism. To make this production-ready, specific code changes are required. The fix should involve modifying the Codex App's MCP client initialization to ensure it's a singleton or session-scoped object. The connection should be established once at the start of a user session or tool-invocation sequence and explicitly closed on session end. The retry-with-backoff mechanism needs to be implemented around the MCP call that might trigger the prompt, specifically handling potential timeouts or errors indicating the user hasn't yet authorized.