Codex CLI repeatedly reports bundled node_repl MCP startup failure on macOS
Open 💬 7 comments Opened Jul 11, 2026 by michycom
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
Summary
Codex CLI on macOS repeatedly prints a startup failure for the bundled node_repl MCP client:
⚠ MCP client for `node_repl` failed to start: MCP startup failed: No such file or directory (os error 2)
⚠ MCP startup incomplete (failed: node_repl)
The session otherwise starts and remains usable, but node_repl is unavailable and the warning recurs.
Environment
- macOS 26.5.2
- Codex CLI 0.144.1
- Terminal: iTerm2
- Working directory: local Xcode/Swift project
- Shell: zsh
Expected behavior
The bundled node_repl MCP client should either:
- start successfully without requiring a manual external path/configuration, or
- remain disabled without emitting a repeated startup failure when its executable/runtime is not present.
If an external dependency is required, the error should identify the missing executable or path and provide an actionable remediation.
Actual behavior
Codex reports only No such file or directory (os error 2), without identifying which file or executable is missing.
Related issue
Possibly related to #31533, although that report concerns Codex Desktop tool exposure, while this report is the Codex terminal CLI failing during MCP startup.
7 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I can confirm I have the same issue: when Codex CLI starts a new session, I receive the following warning messages:
This issue is also completely unrelated to the above issues that the GitHub bot thought to be duplicates.
• I reproduced this with Codex CLI 0.144.4 on macOS 26.4.1
(Darwin 25.4.0 arm64).
I use the standalone Codex CLI and do not have Codex.app installed.
The ChatGPT desktop app wrote the following MCP configuration to the shared
~/.codex/config.toml:``
toml
``[mcp_servers.node_repl]
args = []
command = "/Applications/ChatGPT.app/Contents/Resources/cua_node/bin/node_repl"
startup_timeout_sec = 120
After I uninstalled ChatGPT.app, the standalone Codex CLI continued loading the
persisted entry and attempted to launch the deleted ChatGPT-owned executable:
``
textnode_repl` failed to start: MCP startup failed:MCP client for
No such file or directory (os error 2)
MCP startup incomplete (failed: node_repl)
```
Removing the
[mcp_servers.node_repl]block suppresses the warning.The lifecycle failure is:
node_replexecutablelocated inside its own application bundle.
configuration behind.
on every startup.
No such file or directory (os error 2)withoutidentifying the missing command path or providing remediation.
Uninstalling ChatGPT Desktop should remove or disable the MCP entries it created
for executables contained within its application bundle. Codex CLI should also
handle stale STDIO MCP paths more gracefully by:
config.tomlentry caused the failure; andA separately installed ChatGPT application should not leave standalone Codex
CLI in a persistently degraded startup state after it is uninstalled.
Uploaded diagnostic thread:
019f67f7-ea44-79f2-9b9e-41ff3341f415bump
same here :(
+1 Same issue after uninstalling ChatGPT.app.
Same issue though keep using ChatGPT.app :(