Starting MCP servers (1/2): codex_apps when running /review
Resolved 💬 8 comments Opened Apr 21, 2026 by saadjs Closed May 15, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
What version of Codex CLI is running?
codex-cli 0.121.0
What subscription do you have?
Plus
Which model were you using?
gpt-5.4
What platform is your computer?
Darwin 25.3.0 arm64 arm
What terminal emulator and version are you using (if applicable)?
Ghostty
What issue are you seeing?
When /review is run, CLI is stuck showing Starting MCP servers (1/2): codex_apps but agent is progressing thru (ie. actual review is in process)
What steps can reproduce the bug?
- Uncommitted changes
- Trigger
/review - Select no. 2 (review uncommitted changes)
What is the expected behavior?
_No response_
Additional information
_No response_
8 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
for me those issues started around version 0.120 and never went away. I'm seeing a combination of mcp codex_apps or my custom mcp starting seemingly forewer, while also actually working. And I know it's working because I can see it's output in the main agent. Unless those are startup messages from subagents.
So far, due to various regressions around permissions and now those mcp startup issues, I'm forced to keep using version 0.116 as it's most stable so far.
The problem is that version is said to be supported only up until 5th of may or something.
Can we not EOL for 0.116 just yet please, until you get to the bottom of the latest regressions 🙏🏾 ?
I have also had this persistent problem since about that version. Happy to help debug in any way. I previously had
omxinstalled, but I removed it in case it was the cause of this, including all of its skills and extra folders, to the best of my knowledge. I also agree that I'm skeptical that anything is actually failing to work, and it's just visually incorrect during the run state of mcp_apps (as though the running state is perceived as 'starting'). Either that or the ai works around the issue without the mcp tool.Let me know if I can provide any debugging info.
I should note that my issue happens on more than just
/reviewi think. It feels like it's common during subagent delegation mostly?<img width="1554" height="470" alt="Image" src="https://github.com/user-attachments/assets/39550dca-83ad-48ac-8749-50f527191226" />
attached is how this perma-starting MCP dialog looks like on my end. The session went cleanly, until I asked it to do research via subagents. As soon as it did that, or rather shortly after I saw it started spinning subagents, the MCP loader popped up.
The research itself went fine afaict and the loader eventually ceased, but it doesn't look healthy the way it works. As if something is trying to break free in the backgorund.
The session was started with
--danger-bypassmode or however it's called. That's my default when working in devcontainers.Started yesterday for me, everytime i try to /review it gets stuck forever now
解决了吗
<img width="943" height="538" alt="Image" src="https://github.com/user-attachments/assets/3cfabbf6-85cd-4072-aca8-cd228f090a1d" />
The same issue has been acting up for me on codex v0.130.0 and earlier for the last spawned agent.