Skip to main content
Datasets are the durable outputs of searches, uploads, matches, and workflows. Use Supported MCP tools for exact arguments and terminal states.

​
For AI agents

  • Preserve account_id and dataset_id on every account-scoped call.
  • Use list_datasets, get_dataset, get_dataset_fields, preview_dataset, and get_dataset_lineage for inspection.
  • Show row counts and credit behavior before publishing or downloading.
  • Require explicit confirmation for delete_dataset.
  • Use get_workflow_run when publishing is asynchronous.

​
Inspect a dataset

Call get_dataset, get_dataset_fields, preview_dataset, and get_dataset_lineage as needed. Use the returned schema for later operations. list_datasets supports bounded pagination with limit and offset. Use it to discover dataset IDs, then call get_dataset for metadata and processing runs. get_dataset_lineage can show the parent or child relationship between an upload, match, enrichment, publish, or other workflow output.

​
Publish, download, and delete

publish_dataset starts publishing and may return a workflow run. Poll with get_workflow_run until terminal status. get_download_url is approval-gated when the download has a cost; show the expected row count and cost first. delete_dataset requires explicit confirmation and permanently removes the dataset. Cancellation does not roll back completed work. Publishing may be required before a dataset can be downloaded. After publish_dataset succeeds, use its output dataset ID when requesting a download URL. Download URLs are temporary; treat them as credentials and retrieve the file promptly. MCP returns a URL rather than writing to a local output path, so the client is responsible for saving the downloaded bytes. For deletion, show the target account, dataset ID, and irreversible effect. Call delete_dataset only after explicit approval; do not treat a cancelled delete or workflow as a rollback guarantee.