Skip to main content
Use workflows for enrichment, qualification, research, transformation, publishing, and other asynchronous dataset operations.

​
For AI agents

  • Confirm the account and source dataset_id before starting work.
  • Explain cost and expected output before paid operations.
  • Preserve workflow_run_id and call get_workflow_run before retrying.
  • Call cancel_workflow_run only when the user explicitly asks to stop.
  • Cancellation does not undo rows written or credits spent.

​
Run and monitor a workflow

The starting call may be run_workflow or a specialized workflow tool such as enrich_dataset. Poll the returned run with get_workflow_run. Terminal states are SUCCEEDED, FAILED, and CANCELLED.

​
Specialized workflow operations

Use the specialized MCP tools when they exist so their schemas and approval rules remain explicit: Use the specialized MCP tool that matches the operation and preserve the returned workflow_run_id. If no specialized tool applies, use run_workflow with the supported workflow type.

​
Workflow output and recovery

Successful workflows may produce a new output dataset. Use the output dataset ID from get_workflow_run for the next operation; do not assume the source dataset was modified in place. Terminal states are SUCCEEDED, FAILED, and CANCELLED. Inspect failed or cancelled results before deciding whether a new request is safe.

​
Recovery

A timeout or lost response does not prove that work stopped. Reuse the original workflow run ID. Do not start a duplicate paid operation merely because the client timed out. See Continue a search or workflow.