Plans, quotas and limits

Which plans include the MCP server and the Open API?

The table shows, for each plan, whether it includes each surface, how large its monthly allowances are and how many research calls a minute it allows.

Not in the app yet. Research reports are counted, but the automatic research that writes them is not in the app yet.

How is the monthly allowance counted?

• Every operation call spends one research call, whether it arrives over MCP or over REST: the two surfaces share one counter per account.
• A heavy run — a simulation, a screener backtest, a sweep or a robustness test — also spends one research call, in the same step.
• The allowances recharge at the first instant of each calendar month, in UTC.

What happens when an allowance runs out?

The call is refused with HTTP 429 and the code quota_exhausted; details.resetsAt gives the instant the allowance recharges. A plan that does not include a surface at all is refused with HTTP 403 plan_limit, and details.unlockedBy names the lowest plan that does. Every code is in Errors.

Is there a limit on calls per minute?

• Each connection — an API key, or one AI app connected over MCP — may make the research calls a minute the table gives for its plan, and an account's connections together the per-account figure.
• Every operation call counts, whatever it spends, a refused one included; connecting an AI app and listing its tools do not. The minute is the clock minute, in UTC.
• A call over either figure is refused with HTTP 429 and the code rate_limited before anything runs, so it spends nothing. details.resetsAt gives the instant to repeat it; over REST the Retry-After header gives the same wait in seconds.
• A connection that goes over its own figure never uses up the account's, so one busy script does not stop your AI app.

How large can a single request be?

One request body may hold at most 256 KiB. A larger one is refused with HTTP 413 payload_too_large before anything runs, so it spends nothing.