›_ Codex Issues
Codex Issues / #31762

[Windows] Codex App times out when creating a task or starting a conversation

Resolved 💬 4 comments Opened Jul 9, 2026 by AshYang-right Closed Aug 5, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using (From “About Codex” dialog)?

26.623.141536

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Windows 11 x64

What issue are you seeing?

The Codex desktop app is currently unable to create any new tasks or start new conversations.

Every attempt results in one of the following errors:

  • Error creating task: Timeout
  • Error starting conversation: Timeout

The issue occurs consistently in the Codex desktop app. However, Codex Web works normally with the same account and on the same network connection.

Because both task creation and conversation startup fail before any task begins running, this appears to be specific to the desktop app rather than a repository, prompt, or task execution issue.

Restarting the app does not resolve the problem.

What steps can reproduce the bug?

Feedback ID: no-active-thread-019f4633-e401-7653-8321-9779895d2f93

What is the expected behavior?

The Codex desktop app should successfully create a new task or start a new conversation, as Codex Web currently does with the same account and network connection.

At minimum, the app should return a more specific error message if the request cannot be completed, rather than timing out for every attempt.

Additional information

<img width="282" height="198" alt="Image" src="https://github.com/user-attachments/assets/a32610fb-3c11-4867-af96-47f46122bbff" />

View original on GitHub ↗

4 Comments

github-actions[bot] contributor · 1 month ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #31004
  • #31359
  • #30401

Powered by Codex Action

josheverett · 1 month ago

Update: I fully debugged and resolved my macOS case.

A Codex agent had previously added a local HTTP MCP endpoint as a persistent global entry in ~/.codex/config.toml after I instructed it to use that endpoint. I did not realize the agent had made a durable configuration change.

When that local MCP server later became unhealthy, every new conversation timed out regardless of model, reasoning effort, or speed setting.

Measured behavior while broken:

  • Desktop logged mcp_request_timeout for thread/start after exactly 30 seconds.
  • The app-server returned the corresponding thread/start response only after about 40.1 seconds, too late for the desktop request.
  • mcpServerStatus/list took about 65.1 seconds.
  • App-server initialization and ordinary account/config/model requests completed normally.

Terminating the unhealthy local server immediately restored conversation creation. I then removed its persistent MCP entry and relaunched the app. Afterward:

  • codex mcp get <server> --json reported that the server no longer existed.
  • codex mcp list --json completed in about 4.6 seconds.
  • New conversations worked normally.
  • No further thread/start timeout errors appeared.

The problem began around the time of an app update, but I did not establish whether the update was causally involved. It may have changed MCP startup behavior, exposed an existing configuration hazard, or merely coincided with the failure.

The Codex-side bug is that one unhealthy optional MCP server can delay thread/start until the desktop's generic timeout expires. The app should identify the offending server, mark it unavailable, and continue creating the conversation instead of globally blocking local Codex use.

_Disclosure: this update was drafted and posted by ChatGPT on my behalf from diagnostics I reviewed and approved._

AshYang-right · 1 month ago

I found the root cause on my side.

The timeout was caused by an MCP server that was responding very slowly. Codex was waiting for the MCP server response during task creation, which eventually caused the task creation to fail with a timeout.

After disabling the problematic MCP server, the issue was completely resolved and Codex App started working normally again.

So if you have any custom MCP servers configured, it may be worth temporarily disabling them and checking if the timeout disappears.

Hope this helps!

wherewolf87 · 1 month ago

I encountered the same Windows UI failure and filed native trace evidence in #32799.

The verified failing trace was emitted by client version 26.707.31428. After rebooting into the currently installed 26.707.8479.0, this exact OAuth-lock timeout has not yet recurred in a controlled observation, so current-build recurrence remains unverified.

Observed UI errors included Error creating task: Timeout and Error starting conversation: Timeout. The corresponding native trace showed thread/start -> create_thread -> session_init.mcp_manager_init, followed by six remote MCP clients timing out after 60 seconds while waiting on the same encrypted OAuth secrets-store lock.

That evidence suggests at least one concrete local cause for this otherwise generic UI timeout: optional MCP authentication can block task creation long enough for the Desktop request to become unusable. The report separates this verified lock-contention path from the distinct duplicate local MCP process-pool problem tracked in #32797.