Plans and limits.

Limits apply per workspace and cover writes from the dashboard, the REST API, and MCP alike. Synthetic rows count only rows produced by dataset generation in the current calendar month. Pro is one flat price per workspace, compared in full on pricing.

Every API key is limited to 100 requests per minute. Past that, requests return 429 with the code rate_limited until the window resets. MCP connections authenticate as you rather than as a key, and heavy tool loops should still stay well under the same order of magnitude.

Dataset generation is capped per run on every plan: at most 50 rows per run, and at most 12 rows per run when the dataset has image columns. Larger fills are just multiple runs, each returning its own job id to poll.

When a Free workspace hits a cap, the request fails with status 402 and the code plan_limit. The extra limit field carries one of the limit codes from the table above so your code can tell which cap was hit. Over MCP the same failure comes back as a tool error whose text starts with plan_limit:.

{
  "error": {
    "code": "plan_limit",
    "limit": "dataset_rows",
    "message": "The free plan caps datasets at 100 rows. Upgrade to Pro for unlimited rows."
  }
}