ChatGPT Scheduled Task Says a Connector or App Is Unavailable
If a connected app works in a normal ChatGPT conversation but becomes unavailable when a Scheduled Task runs, test the task’s runtime—not just the connector. The scheduled run may lack the live chat’s app selection, account context, project files, or ability to obtain approval for a consequential action.
The useful distinction is whether the task cannot find the connector, cannot authorize it, or finds it but cannot complete the action unattended.
Start with the same read-only request in both places
Run this in a normal chat:
Using the connected Gmail app, list the subjects of the three newest unread emails. Do not change anything.
Then schedule the same read-only request for a few minutes later, including the app name and expected account.
- If both fail, repair the connector.
- If live chat works but the scheduled run cannot find the app, the app is unavailable in that task context.
- If both read successfully but the scheduled write fails, the issue is probably approval or write permission.
Do not test with sending, deleting, purchasing, or updating first. Those actions add a second failure layer.
Check the connected account and workspace
Apps are scoped to the ChatGPT account or managed workspace where they were authorized. Confirm that:
- the task was created under the same ChatGPT account;
- the app is enabled in that workspace;
- the external account is still authorized;
- the workspace administrator has not restricted the app or its write actions;
- the task names the intended app and external account explicitly.
Account switching keeps chats, workspaces, and connected apps separate. A connector working in your personal account says nothing about a task running in your work workspace.
Remove project-file and GPT dependencies
OpenAI says a Scheduled Task created in a project cannot access that project’s files. Voice chats and GPTs are also unsupported in Scheduled Tasks.
If the live version relies on a file or custom GPT, the scheduled version needs another accessible source. Put the required data in a supported connected app, or rewrite the task so it can run from information available at execution time.
Look for an approval dead end
Apps may require approval before they change external data. In a live chat, ChatGPT displays the proposed action and you approve it. An unattended scheduled run cannot always pause mid-workflow, collect that decision, and continue as if you were present.
Use one of three designs:
- Read and report only: the task creates a review queue.
- Predefined workflow elsewhere: an automation system executes already-authorized actions.
- Human-in-the-loop: the task notifies you and you complete the write action in a live session.
If the job must send an email automatically, “approval required every time” is not a prompt problem. It is an execution-model mismatch.
Reconnect and refresh the app catalog
For a stale OAuth connection:
- Disconnect the app from ChatGPT.
- Revoke the old authorization in the external service if necessary.
- Reconnect the intended account.
- Test one read operation in a new chat.
- Recreate the Scheduled Task rather than relying on an old task context.
For custom MCP connectors, verify that the server advertises valid tools and is reachable when the task runs. If you see api_tool.list_resources returning nothing, follow the api_tool and list_resources troubleshooting guide.
Make every dependency explicit
Too vague:
Check the CRM every morning and handle new leads.
Testable:
Every weekday at 8:00 AM America/New_York, use the connected HubSpot app for the Example Company workspace. Find contacts created in the previous 24 hours and return name, company, source, owner, and record URL. Do not create or update records.
Once the scheduled read succeeds, decide separately how the approved follow-up should execute.
Multiple accounts can look like a missing connector
The app may be available while the requested data belongs to a different Gmail, Outlook, calendar, or CRM identity. Name the expected account in the task and confirm it in the output.
When one workflow needs several inboxes, Carly can hold each authorized connection separately and route actions by full account address. See multiple Gmail accounts in ChatGPT Work and multiple Outlook accounts in ChatGPT Work.
Use Carly when the connector must act unattended
Carly’s workflows run outside the ChatGPT Scheduled Task runtime. You connect the external accounts, define the trigger and allowed actions, test the workflow, and let it execute in the cloud. ChatGPT can remain the place where you investigate or design the job without becoming the scheduler that must also negotiate every app approval.
That is the better architecture for “when a lead arrives, enrich it, update the CRM, send the reply, and create the follow-up.” Scheduled Tasks is better for “every morning, tell me which leads arrived.”
Frequently asked questions
Why does the connector work manually but not on schedule?
The live conversation may have an enabled app, file, account context, or approval path that is unavailable to the scheduled runtime. Reproduce the task with one read-only app call and add dependencies gradually.
Can a Scheduled Task use project files?
OpenAI says a task created in a project cannot access that project’s files. Move the required information to a supported connected source or remove the dependency.
Can I preapprove every ChatGPT app write?
Approval behavior depends on the app, action, plan, and workspace policy. Do not assume a write can run unattended simply because it works after approval in a live conversation.
Related: ChatGPT Scheduled Tasks guide · Scheduled Tasks not working · api_tool and list_resources errors · ChatGPT connectors · Carly integrations
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
