Rate limits
Each API key may make a limited number of requests per minute. The limits stop one runaway script from slowing the API down for everyone. They are separate from credits, which cap how much work you can run.
| Tier | Limit | Endpoints |
|---|---|---|
| Expensive | 30 requests per minute | POST /v1/responses |
| Read | 120 requests per minute | Every other endpoint |
- One tier per request. A request counts against its own endpoint’s tier only, so starting turns never uses up your read budget.
- Per key. Each API key has its own budget. A signed-in MCP client (Claude, ChatGPT) has one budget per client and account.
- Fixed one-minute windows. The count resets at the start of each minute.
Headers
Section titled “Headers”Every rate-limited response, success or error, carries the state of its tier:
| Header | Meaning |
|---|---|
RateLimit-Limit |
Requests allowed per window |
RateLimit-Remaining |
Requests left in the current window |
RateLimit-Reset |
Seconds until the window resets |
Over the limit
Section titled “Over the limit”A request over the limit is refused with 429 and the code rate_limited. Its Retry-After header gives the seconds until the next window. Wait that long and send the request again.
Polling a background response every few seconds stays well inside the read tier. If you need higher limits, contact us at marvin@marvin-labs.com.