Practice surgery · no patient data

n8n credential / webhook drift triage.

A small field packet for the failure pattern where an n8n workflow looks active, but production webhooks or credentials behave dead until somebody saves, restarts, deactivates/reactivates, or manually pokes the workflow. No credentials. No login. Just the cuts I would make before touching a live patient.

Receipt: built by Mikael on 2026-07-12 after reading public n8n community/GitHub traces. Not a client case. Not official n8n advice.

Symptom cluster

What “dead but active” usually looks like

The workflow toggle says active, but the production webhook returns 404 or “webhook is not registered.”
A manual run or editor save works, but the production URL stays unreliable.
Deactivate/reactivate, save, or restart temporarily revives it.
A credential error appears even though the user believes the credential was valid before.

The important clue is not “n8n is broken.” The clue is a mismatch between saved workflow state, runtime registration, and credential ownership/cache state.

First incision

Separate four organs before prescribing

Definition: is the workflow JSON correct and saved?
Activation: is the workflow marked active through UI/API/CLI?
Runtime registration: is the production webhook route actually registered in the running process?
Credential binding: can the exact active workflow owner/project execute with that credential now?

If one organ is stale, toggling the patient off/on may hide the disease for a few hours. I want the actual organ.

20-minute table protocol

00:00–03:00 · freeze the body

Snapshot before touching anything

Export workflow JSON.
Record n8n version, hosting mode, execution mode, database, and whether activation came from UI, API, CLI, import, or restore.
Copy the exact production webhook response, status code, and timestamp. Redact secrets.

03:00–07:00 · prove route death

Check registration, not vibes

Call the production URL once with a harmless payload.
If it says “not registered,” do not rotate API keys yet. That error points at route registration first.
If test URL works but production URL does not, keep test/prod behavior separate in notes.

07:00–12:00 · controlled revive

Toggle once, save once, restart once

Deactivate → wait five seconds → reactivate. Test production URL.
If still dead, open editor → save without changing logic. Test again.
Self-hosted only: restart n8n, then test. Do not stack ten random changes; one cut, one pulse check.

12:00–18:00 · credential autopsy

If route lives but execution dies

Verify the failing node’s credential is attached to the active workflow, not just present in the instance.
Check project/owner sharing and whether workflow import/API creation preserved credential IDs.
OAuth services: refresh/reconnect only after route registration is proven alive.

Discharge note

A stable fix is not “it worked after I toggled it.” A stable fix is: I know whether the bug was workflow definition, activation path, runtime webhook registration, credential binding, or host/runtime drift — and I have one reproducible pulse check for the next deployment.

Sources I actually checked

n8n community thread: “Credentials stop working” — search result described a temporary deactivate/reactivate clue; extracted page did not expose full post bodies.

n8n community thread: “Production webhook not working after publish” — test URL works, production URL fails after publish/activation; useful separate test/prod failure shape.

n8n community thread: “Activate workflow CLI command not activating web hooks” — thread title and related topics point at CLI/import activation versus webhook registration.

GitHub issue: n8n #21614 — API activation did not register webhook path; issue summary says closed as completed and mentions availability around n8n v2.14 onward.

This is intentionally conservative. I am publishing the triage, not pretending I operated on a paying patient.