/api/v1 burst responses and error results from the six
existing COLA Cloud MCP tools. Limits, monthly allowances and successful tool
schemas are unchanged. No data refresh or historical reprocessing is involved.
The API distinguishes a rate-limiter service failure (503 limiter_unavailable)
from an enforced burst limit (429 rate_limit_exceeded). Both continue to deny
the request. Burst Retry-After and error.details.retry_after report the
seconds until the next minute window, rounded up, instead of a constant 60.
Previously, MCP dropped retry/reset timing and suggested calling get_plan on
every 429. MCP now preserves bounded timing and meter details, advises waiting
before any next tool after burst rejection, and advises pausing monthly-exhausted
meters until reset or a confirmed entitlement change. It performs no automatic
retries and adds no billing links or tools.
Your action: distinguish error.code in your caller. Honor Retry-After for
burst or service-unavailable responses with a finite retry budget. Persist a
pause for quota_exceeded until X-Quota-Reset or confirmed entitlement change.
Do not immediately call /usage or get_plan after a burst rejection.
Reproducible examples: a burst response with Retry-After: 23 becomes an MCP
error with retry_after_seconds: 23; a monthly error with
X-Quota-Reset: 1793491200 carries quota_reset_at: 2026-11-01T00:00:00.000Z
and retryable: false. This timestamp is an example, not the current reset.
See authentication and recovery guidance for error codes and
request corrections, and MCP release #15
for the Worker implementation.