Tools troubleshooting: symptoms and recovery
Maps Tools symptoms like unavailable catalogs, blocked calls, and paused Tools to their first checks and safe recovery actions.
This guide covers symptom diagnosis, tool pause recovery, unknown outcome handling, connection recovery, and support escalation for Tools.
- Symptom-to-check mapping for catalog status and call outcomes.
- Tool pause behavior and recovery testing procedure.
- Unknown outcome reconciliation and idempotency guidance.
- Connection recovery without recreation.
- Support escalation details and required information.
Summaries were generated by AI.
Start from what the Dashboard shows under Tools. The Dashboard does not show why a catalog check failed, so each branch below checks what you control: the endpoint, the remote credential, the call policy, and the assignment. Operate Tools defines every state and outcome named here. For an SDK Tool that your own process serves, use the Tool Host section of SDK troubleshooting instead.
Find the symptom
A Tool is temporarily paused
Duale AI paused calls to that Tool on that Connection after repeated technical failures, and did not send this call to the server. Understand the MCP call contract states the pause rule and its length. Catalog checks and other Tools on the Connection continue during the pause.
Wait until the pause ends, then test the Tool once with a low-risk Task. Do not send repeated Tasks to force recovery.
Check. The test call ends as Succeeded in the Task view or in Analytics. If it pauses again, check the server’s own logs for the failing requests before you test again.
An outcome is unknown
Duale AI cannot prove the call’s result, and it cannot rule out that the server received the call. The remote operation can have run, or not.
- Read the destination system’s own record of the operation.
- If the operation happened, record it and continue from that state.
- If it did not happen, repeat it only when a repeat is safe.
Make state-changing remote operations idempotent so that a repeat cannot double an effect. Secure integration gives the procedure and the identifiers to keep.
Check. The destination system shows the operation exactly once, or not at all. Stop if you cannot read that record; do not repeat the operation.
Recover without recreating the Connection
When a server is down, Duale AI keeps checking until you disable or delete the Connection. Correct the endpoint or the remote credential: saving the change starts a new check at once. After you fix the server itself, let the next scheduled check refresh the catalog. Deleting and recreating the Connection does not speed recovery, and you must assign the new Connection again.
Check. The catalog shows Active with a recent last successful check.
Ask for support
If the endpoint and credential appear correct and the state does not recover, email technical support. Include the Connection ID, Connection name, current status, approximate time of the problem, and the last successful check time if the Dashboard shows one. Never send the bearer token, API key, or unrestricted logs by email.