Request limits

Each JSON-RPC request counts one unit in one-minute windows. Ordinary requests allow 240 per person and 1,200 per organization. Portal session calls, audio transcription and web search or fetch have separate limits of 20 per person and 60 per organization. Requests without a valid token allow 60 per IP.

Responses include RateLimit-Policy, such as "mcp-person-standard";q=240;w=60, and RateLimit, such as "mcp-person-standard";r=239;t=45. q is the quota, w the window in seconds, r the remaining requests and t the seconds until the next window. The headers report the most constrained limit, following the IETF draft, revision 11.

When a limit is reached, the response is HTTP 429 with Retry-After in seconds and JSON-RPC error -32029. The request id is echoed when parseable. error.data.error has code: "rate_limited", retryable: true, scope of person, organization or ip, class of standard or expensive, and retryAfterSeconds. Wait that long before retrying. If the counter store is unavailable, requests continue without quota headers.