Troubleshooting

The errors you'll actually hit, in plain language, with the fix.

The errors you'll actually hit, in plain language, with the fix. Build & Publish's diagnosis dialog links straight here — each heading matches a diagnosis rule.

"Log Search required" / Tracer looks thin

Symptom: an orange banner in Tracer's task list mentioning Log Search, an Advanced package, or Titanium — and fewer tasks than you expect, with no trace spans.

Cause: full-fidelity tracing needs Enhanced Log Search (the Advanced package or a Titanium subscription). Without it, Tracer degrades to runtime logs.

Fix

Nothing is broken — this is reduced fidelity. You still get tasks and details from runtime logs (capped at 1000 lines, no spans). For the full experience, have an org admin add the entitlement. See the entitlement section.

Target locked after first deploy

Symptom: a deploy fails with a wall of "component aborted" / errorCode 3025 lines, and text like "target cannot be changed after the initial deployment."

Cause: Anypoint locks a network's deployment target after the first successful deploy. The 3025 cascade is a symptom; the real cause is a single 400 on the deployment update. This is the most common deploy failure.

Fix

Deploy back to the original target — or delete the deployment in Runtime Manager and deploy fresh to the new gateway/space.

Broker can't reach its connections

Symptom: deploy fails with "unreachable connections" or "failed connection validation" (sometimes with a 404).

Cause: at deploy time the broker validates it can reach the endpoints it depends on (MCP servers, agents). Runtime Manager holds it pending until the CLI times out.

Fix

Deploy the connection's target first (e.g. the MCP server), check its URL/port/path, and make sure the broker and its targets share a gateway/space. A 404 means nothing is served at that address.

Deploy timed out

Symptom: the deploy step runs to its timeout (5 min default) and is terminated.

Cause: a deploy rarely hangs at random. Almost always the app was created but failed to start, or failed health/connection validation.

Fix

Read the Runtime Manager log that Build & Publish auto-fetches into the diagnosis dialog — it says why the app didn't go healthy. Raising LIFECYCLE_DEPLOY_TIMEOUT_MSwon't help if validation is what's failing.

Permission denied

Symptom: a 403 / "forbidden" on publish, deploy, unpublish, or undeploy.

Cause: a role or OAuth scope gap — not a project problem. The CLI runs as you, so your own grants govern the action in that business group.

Fix

Ask an org admin to grant the role for this action in that org, then use "Refresh Anypoint permissions" from the header account menu to pick up the new grants.

No build artifacts (errorCode 3046)

Symptom: deploy fails complaining there are no build artifacts.

Cause: deploy needs a build's target/ output, and it wasn't present — normally the worker chains build in automatically.

Fix

Resubmit the deploy (it re-runs build first). If it recurs, the build step itself is failing — check the earlier build output in the log.

Can't unpublish (active instances)

Symptom: unpublish is refused because active API instances exist.

Fix

Undeploy first, then unpublish. Anypoint won't erase an Exchange version that still has live instances. Expect to revoke any active API contracts during undeploy.

Hard-delete window

Symptom: a hard delete is rejected mentioning a hard-delete restriction.

Cause: Anypoint only permits hard delete roughly 7 days after asset creation, and an org can disable it entirely.

Fix

Wait out the window, or have the org enable hard delete. As a last resort use soft delete — but remember it burns the version number permanently. See teardown.

"Converting circular structure to JSON"

Symptom: a cryptic "Converting circular structure to JSON" error.

Cause: the CLI (oclif) couldn't serialize its own error object — almost always because it was passed a flag the installed plugin version doesn't support. The real cause never reaches the log.

Fix

This is a server-side toolchain issue, not something you did. If you self-host, check the CLI/plugin versions in Ops; otherwise report it — the diagnosis dialog flags it as a rejected flag.

Tracer shows nothing

Symptom: a broker is deployed, but the task list is empty or a task's Task story / LLM reasoning panels are blank.

Causes & fixes:

Wrong scope — confirm business group → environment → broker, and widen the activity period.
Object Store expired — Task story / reasoning age out (24h default TTL). An old task simply won't have them anymore.
Monitoring not enabled — the deployment needs the INSECURE-LOGGING category enabled in Runtime Manager for full task visibility.
Entitlement — see above; without Log Search you see fewer tasks.
Just deployed — no one has invoked the broker yet, so there are no tasks. Use Tracer's Invoke rail to create one.

Builder won't export

Symptom: "Save to folder" or "Download .zip" throws"Project validation failed" or "AgentScript conformance failed".

Cause: file exports are hard-gated on zero validation errors and clean AgentScript conformance.

Fix

Open the validation strip and click each blocking issue — it jumps to the exact field. Clear them all (the green "Valid" chip), then export. To park a work-in-progress, use Save in browser, which has no gate. See saving & exporting.

Still stuck? Use the bug button in the header to send a report with a screenshot.