Pricing

Codex "Error creating task" (Timeout): Why Cloud Tasks Fail and How to Fix It

Codex stuck on "Error creating task" or a timeout? Why cloud tasks fail to start, fixes reported in OpenAI's GitHub issues, and how to check if Codex is down.

By Engineer Foodie Desk Nov 3, 2026 ✓ Sources checked Nov 3, 2026

8 min read

Error wording checked Nov 3, 2026 in four issues on OpenAI’s official Codex GitHub repo (#32861, #31762, #9001, #8022). MCP settings checked Nov 3, 2026 in OpenAI’s Codex MCP docs, and cloud setup in OpenAI’s Codex cloud docs. Plan eligibility checked Nov 1, 2026 in OpenAI Help. Every source is listed at the bottom. No affiliate links on this page (how we work).

Here’s the finding up front. In the GitHub reports I read, Error creating task Timeout was usually not a cloud outage. It was usually Codex hanging while it started a new thread, and the most common thing it hung on was a slow or broken MCP server in your own config. The second biggest group was a bad local config file. Actual cloud-task problems are a smaller, separate pile.

OpenAI hasn’t published a doc entry for this error as of Nov 3, 2026. So everything below comes from issue threads, with each fix labeled by who reported it.

The exact error: `Error creating task Timeout`

This is the wording from the title of #32861, opened July 13, 2026 on the macOS Codex app (Plus plan):

Error creating task Timeout

The Windows report #31762 shows it with a colon, plus a twin message that turns up next to it:

Error creating task: Timeout
Error starting conversation: Timeout

The VS Code extension version from #9001 leaves the timeout off: Error creating task or Error starting conversation.

One-line answer: Codex couldn’t start a new thread within its time limit. Check your MCP servers first, then your local config files, then whether you meant to start a cloud task at all.

The #32861 reporter’s steps say a lot: “Hit New on a folder in codex”, type a prompt, “Get a timeout”. That’s a local task on a folder, not a cloud job. So “task” here means “a new Codex thread”, wherever it runs.

Cloud task vs local task: which one is failing?

This matters, because the fixes are different.

Local (desktop app or IDE extension, on a folder): this is where #32861, #31762, #9001 and #8022 all come from. The #31762 reporter said “Codex Web works normally with the same account and on the same network connection”. That points away from your account and toward the local app.
Cloud (Codex on the web, or a task sent to a cloud environment): this needs a published cloud environment with a connected GitHub repo. OpenAI’s docs walk through it: pick Cloud, select or create an environment, “Connect GitHub if prompted”, publish, and wait for “Environment published” (Codex cloud docs).

The quick test: if Codex works on the web but not in the app, the problem is local. If it fails everywhere, check status.openai.com and your cloud environment.

In the IDE extension you can start a cloud task without meaning to. In #9001, OpenAI contributor etraut-openai asked the reporter: “are you trying to start a cloud task?” He pointed to the selector “In the lower left part of the extension panel” and said: “Unless you intend to start a cloud task, make sure this is set to “Local Workspace”.”

How to fix Codex “Error creating task”

In order, cheapest first.

1. Find the MCP server that’s holding up thread start (reported in #31762 and #32861)

This is the most common fix in the threads, and it’s very Lester Freamon. All the pieces matter, especially the one you installed and forgot about.

In #31762, a user found a local HTTP MCP endpoint in ~/.codex/config.toml that had gone unhealthy. The desktop app “logged mcp_request_timeout for thread/start after exactly 30 seconds”. Shutting down that server “immediately restored conversation creation”.
Another #31762 user: “The timeout was caused by an MCP server that was responding very slowly.” Disabling it “completely resolved” the problem.
In #32861, a commenter said theirs was “related to an MCP that I’d installed that was being called at the beginning of all thread launches”.
A third #31762 commenter traced the hang to session_init.mcp_manager_init, with remote MCP clients stuck waiting on a lock for OAuth credentials.

How to do it with documented tools (Codex MCP docs):

Run codex mcp list to see every configured server. In the terminal UI, /mcp shows the active ones.
Disable suspects one at a time without deleting them. The docs say enabled: “Set false to disable a server without deleting it.”
Look for required = true. Per the docs, a required server that can’t start makes Codex startup fail.
Once you find the slow one, its startup_timeout_sec (default 10) and tool_timeout_sec (default 60) are the documented knobs. Fixing or removing the server is the real fix, though.

2. Check for broken config files (reported in #9001)

In #9001, the reporter ran Codex CLI 0.80, which showed the error the extension had been hiding: ~/.codex/skills/.../SKILL.md: invalid YAML. Fixing that file fixed the extension. Another commenter’s cause was invalid TOML in ~/.codex/agents/*.toml.

OpenAI’s etraut-openai replied: “Codex should be more resilient here — or at least provide a better error message.” I agree. One typo in a skill file shouldn’t brick the whole extension behind a vague message. Running the CLI to read your own error messages is a funny kind of support ticket.

The practical move is to open a terminal and run codex in the same folder. The CLI may name the broken file. If it does, fix or move that file and try again.

3. Restart or reinstall the app or extension (reported in #9001 and #8022)

etraut-openai in #9001: “Did you try restarting the Codex IDE extension?”
“Vscode restart helped” (#8022, macOS).
Reinstalling the IDE extension worked for one #9001 commenter.
Caveat: the #31762 reporter said “Restarting the app does not resolve the problem.” A restart doesn’t fix the MCP cause.

OpenAI’s troubleshooting page also notes that the desktop app and the CLI can ship “different Codex versions” (troubleshooting). Check codex --version if the CLI works but the app doesn’t.

4. Start a new chat instead of resuming an old one (reported in #8022)

Several #8022 comments say the error shows up when resuming or prompting an old chat. One said it “doesn’t happen in new ones”. OpenAI’s troubleshooting page also suggests, for stuck chats, “Start a new chat with a smaller, more focused prompt”. This is a workaround, not a fix, and it’s the issue that’s been open since Dec 14, 2025. Time is a flat circle, and in this thread the circle is a spinner.

5. For real cloud tasks: fix the environment (per docs, and reported in #9001)

In the environment, publish it and wait for “Environment published”. If setup fails, OpenAI’s docs say to “select Try again”, then publish or republish after fixing it (cloud environments).
Check repo access. One #9001 commenter’s repo had lost authorization on GitHub (HTTP 403 in the GitHub output). The docs say “Repository access and personal connections depend on the account running the task.”
Codex cloud doesn’t support GitLab or self-hosted GitHub Enterprise Server, per the same page.
Codex cloud comes with Plus, all Pro tiers, Business, Enterprise and Edu (OpenAI Help, verified Nov 1, 2026).

6. Send feedback with logs

etraut-openai asked #9001 reporters to use /feedback and post the thread ID. In the desktop app, type / in the composer to send feedback and get a session ID. On macOS, logs live in ~/Library/Logs/com.openai.codex/YYYY/MM/DD (troubleshooting). The mcp_request_timeout line from #31762 is exactly the kind of clue that turns up there.

“Error creating chat” timeout

The desktop app has a sibling: a red toast reading Error creating chat. In #34937 (Windows, Pro) it came with failed to start turn: internal error; agent loop died unexpectedly, followed by “Error submitting message” on every message “until the app is restarted”. Searching OpenAI’s repo turns up 75 issues with “Error creating chat” as of Nov 3, 2026. I couldn’t find an official doc or OpenAI fix for a timeout version of it. Reporters recovered by restarting the app. That message-side failure is covered in Codex error submitting message.

Which fix matches your symptom

Symptom	Likely cause	Try
Web works, app fails	MCP server	Disable MCPs
Extension fails only	Bad skill file	Run CLI
Old chat fails only	Resumed thread	New chat
Cloud task only	Environment	Republish
Fails everywhere	OpenAI side	Check status

Here’s what I’d do

Run codex mcp list. If you added an MCP server around the time this started, set enabled = false on it and try again before anything else. That’s the fix with the most reports behind it. If that’s not it, run codex in a terminal in the same folder and read what it complains about. Then restart, then try a fresh chat. If Codex Web also fails, stop debugging your laptop and check the status page.

What bugs me is that the message says “Timeout” and nothing more. Codex knows which MCP server it was waiting on, because the logs show it. Putting the server name in the toast would close half of these threads.

If timeouts keep eating your workday, that’s a reasonable moment to look at Codex alternatives or the wider AI coding tools compared page. For other Codex failures, see Codex stream disconnected for mid-response drops and Codex usage limits for quota questions.

FAQ

Why does Codex say Error creating task Timeout? Codex couldn’t start a new thread in time, and GitHub reports most often trace that to a slow or broken MCP server in your config.

Is Codex down? Check status.openai.com; if Codex Web works on the same account and network, the problem is likely your local app or config, not an outage.

Does a failed task count against Codex usage? Not published; OpenAI’s help pages don’t say whether a task that fails to start uses any of your Codex limits.

How do I disable an MCP server in Codex? Set enabled = false for that server in your Codex config, which OpenAI’s MCP docs say disables it without deleting it.

Does restarting the Codex app fix Error creating task? Sometimes; restarts helped in some reports, but one Windows reporter said restarting did not fix it when an MCP server was the cause.

Sources

OpenAI Codex GitHub issue #32861, “Unable to use Codex in new ChatGPT app, always get Error creating task Timeout”: https://github.com/openai/codex/issues/32861 (verified 2026-11-03)
OpenAI Codex GitHub issue #31762, “[Windows] Codex App times out when creating a task or starting a conversation”: https://github.com/openai/codex/issues/31762 (verified 2026-11-03)
OpenAI Codex GitHub issue #9001, “Error creating task or Error starting conversation”: https://github.com/openai/codex/issues/9001 (verified 2026-11-03)
OpenAI Codex GitHub issue #8022, “Error Creating Task”: https://github.com/openai/codex/issues/8022 (verified 2026-11-03)
OpenAI Codex GitHub issue #34937 (“Error creating chat” toast): https://github.com/openai/codex/issues/34937 (verified 2026-11-03)
OpenAI Codex docs, MCP (codex mcp list, enabled, timeouts, required): https://learn.chatgpt.com/docs/extend/mcp.md (verified 2026-11-03)
OpenAI Codex docs, troubleshooting (feedback, logs, versions, stuck chats): https://learn.chatgpt.com/docs/reference/troubleshooting.md (verified 2026-11-03)
OpenAI Codex docs, cloud: https://learn.chatgpt.com/docs/cloud.md (verified 2026-11-03)
OpenAI Codex docs, cloud environments: https://learn.chatgpt.com/docs/environments/cloud-environments.md (verified 2026-11-03)
OpenAI Help, using Codex with your ChatGPT plan (cloud plan eligibility): https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan (verified 2026-11-01)
OpenAI status page: https://status.openai.com (checked 2026-11-03)

The exact error: Error creating task Timeout