- Verify whether the Codex restart plus host reconnect attached the live MCP client to the updated hot shell.
- Determine whether the previously observed long-call defect still exists after the restart.
- Separate
DevServerfailures from general shell or worker health.
Clipboard(get): passSnapshot: passWait(1): pass- Interpretation:
- the live shell is not broadly wedged;
- the MCP connection is active and normal public tools still work.
- Expected:
- pass if the restarted shell loaded the shell-local
Waitmitigation.
- pass if the restarted shell loaded the shell-local
- Actual:
- pass.
- Interpretation:
- the restarted live shell is serving the patched shell-owned
Waitpath.
- the restarted live shell is serving the patched shell-owned
- Expected:
- prompt JSON health response from the restarted shell.
- Actual:
- still timed out at the outer 120 second tools/call deadline.
- Contrast evidence:
- live
Clipboard,Snapshot, andWaitall succeeded; - local shell
DevServer.healthreturned immediately with the new source-hash fields.
- live
- Interpretation:
- this is now a tool-path-specific live integration failure, not evidence that the shell restart failed.
Start-Sleep -Seconds 8; Write-Output 'wmcp-long-call-ok': fail, outer 120 second timeoutStart-Sleep -Seconds 4; Write-Output 'wmcp-short-call-ok': pass- Interpretation:
- the long-call defect is generic to forwarded worker calls and not specific to
Wait.
- the long-call defect is generic to forwarded worker calls and not specific to
- Live process tree observed through PowerShell:
uv.exe --directory C:\Users\Rufus\src\Windows-MCP run windows-mcp --dev hot- child
windows-mcp.exe - child shell
python.exe - child worker
python.exe -m windows_mcp.__main__ --internal-worker --transport streamable-http --host 127.0.0.1 --port 49946 --generation 1
- The host reconnect succeeded and the live shell restart took effect for shell-local behavior.
- The remaining reliability defect is the internal worker
streamable-httpbridge for longer-running forwarded calls. DevServershould now be treated as a separate host-facing integration problem, not as the main indicator of shell freshness.