Skip to main content
A run collects answers for the selected prompts and models immediately instead of waiting for the monitoring schedule. Read MCP write access first for plans, scopes, roles and confirmation tokens. Examples use synthetic data.

start_run

Collect answers now for the selected prompts and models. The tool reserves the credits shown in the preview, so it needs a confirmation token on every execution.

Parameters

Before you start

  1. Call estimate_run to see the usable prompts, models and the estimated credits.
  2. Call get_credits if you need to check that the balance covers the cost.
  3. Call start_run without a token to receive the preview and the token, then repeat the call with the token.

After the run starts

Collection is asynchronous. Follow progress with get_run and list_runs. There is no MCP tool to cancel or retry a run. To be notified by your own systems when a run finishes, subscribe to the run.completed event with the REST API.
The preview shows the credits that would be reserved. Moving credits between teams and buying more remain dashboard actions.

Example