Skip to main content
HTTP failures return an error object with a stable code and a message. Optional details carry recovery information when available. MCP tool failures set isError and return the shared error information in text content. Treat an error separately from a successful check whose observation is unknown.
This is an illustrative scoped-access refusal. Keep the code and relevant identifiers in diagnostics, while keeping credentials out of logs. Server failures use a public message and withhold private error details. Configuration failures can return retryable: false and action: operator_configuration_required.

Access and input

See HTTP setup and MCP setup. Access errors need a corrected identity or grant; repeated identical requests cannot widen permissions.

Checks, limits and replay

A 429 can describe a request-rate limit or an exhausted allowance. Read its code before scheduling a retry. See usage, checking a placement and batch verification.

Expired history and configuration

CURSOR_EXPIRED returns 410 when a cursor falls outside retained history. Follow its recovery details. A source-event refusal provides resync_required, snapshot_endpoint and resume_cursor; export the current snapshot before resuming. Keep destination and source feeds separate. A missing historical interval cannot be reported as no changes. EVIDENCE_EXPIRED returns 410 when the raw observation snapshot has expired. Observation metadata remains available. A new check produces current evidence, not a recreation of the old page. PROVIDER_NOT_CONFIGURED, PROVIDER_DISABLED, OWNED_DISCOVERY_NOT_CONFIGURED and OWNED_CORPUS_UNAVAILABLE require operator configuration. Repeated calls cannot activate a provider or load a corpus. Use an available customer import when it fits the task. Consult the operation’s page for its remaining domain-specific errors. Troubleshooting gives end-to-end recovery; common schemas explains result fields and uncertainty. Availability distinguishes source contracts from your connected release.