Quick answer

Startup plan gets 300 requests per minute. Enterprise gets 500. Limits are per account, shared across all your API tokens. Exceeding returns HTTP 429 Too Many Requests with a Retry-After header. Every response includes X-RateLimit-Remaining and X-RateLimit-Reset so you can back off before hitting the limit.

Response headers

Every API response includes:

Use these to back off preemptively. A well-behaved client keeps Remaining above 20 and slows down as it approaches zero.

Handling 429 gracefully

if response.status_code == 429:
    wait_seconds = int(response.headers.get("Retry-After", "60"))
    time.sleep(wait_seconds)
    # retry

Don't retry without waiting. Don't retry more than 3 times. Don't retry non-idempotent operations (POST/PATCH) without checking whether the first request actually succeeded.

Best practices for bulk work

Upgrading for higher limits

The main reason to upgrade to Enterprise if you're already at ~200 mailboxes — the 500 req/min limit is enough for most AI SDR pipelines running full-tilt. Beyond that, contact support about custom rate limits.

What's next

Frequently asked questions

What's the burst allowance?

There's headroom above the steady per-minute rate. Startup absorbs up to 500 requests within a minute-window, Enterprise up to 800, before the limiter kicks in. Sustained sending above your plan's rate gets throttled.

Are heavy endpoints (export) rate-limited differently?

Export generates a downloadable file and is treated as one request regardless of how many mailboxes it includes.

What happens when I hit 429?

Server returns 429 with a Retry-After header (in seconds). Wait that long, then retry. Aggressive retry loops without respecting Retry-After will get you rate-limited longer.

Are jobs rate-limited?

The API call to create a job counts against your rate limit. The job itself running in the background doesn't. So provisioning 100 mailboxes = 100 POST calls (~20 seconds on Startup at 5/sec), plus background processing time (unaffected by rate limits).

Do MCP tools count against the rate limit?

Yes — MCP calls Winnr's API under the hood, so each MCP tool invocation is at least one API request. Agents that hammer many tools quickly can hit rate limits.

Was this article helpful? Yes · No