Inboundr meters usage per account, not per key or per endpoint. Every authenticated call — a REST request or an MCP tool invocation — draws from the same budget:
5 requests per second, per account. The REST API and the MCP server share one bucket, keyed to the owner of the API key. Extra keys don’t buy extra throughput — they all spend from the same account budget.
The window is a rolling one-second counter, so your allowance refreshes every second rather than draining toward a distant monthly reset.

​
Reading your budget on every response

You never have to guess where you stand — each response carries your current state: When a call is rejected you also get a retry-after header (in seconds) and a 429 with:

​
Staying under the limit

Because the budget is shared, the thing to watch isn’t any single loop — it’s everything hitting Inboundr on your account at once.
An agent working through an inbox can fire tools quickly. Sequence its list_emails → get_thread calls rather than fanning them out in parallel, and it will comfortably stay within five per second.
Reading a large mailbox? Raise limit (up to 100) and paginate instead of firing many small, concurrent list calls.
On a 429, wait the number of seconds in retry-after before retrying, and back off further if you keep hitting the wall. Since the window is one second, a brief pause is usually all it takes.
Building something that legitimately needs more headroom? Reach out — higher per-account limits are available for high-volume workloads.