Skip to main content
MCP responses are structured objects, not terminal output. Exact optional fields can vary by tool and operation, so use the returned schema and preserve identifiers exactly.

​
For AI agents

  • Preserve account_id, dataset_id, search_handle, agent_run_id, and workflow_run_id.
  • Do not claim completion from a start response; inspect the terminal status.
  • Use get_dataset_fields for the current dataset schema.
  • Keep error and confirmation objects visible to the user.

​
Search response

Use get_search_result for a returned search handle. If an agent_run_id is present, use get_agent_run or wait_for_agent_run and use the list/latest tools when the trigger response was lost.

​
Dataset and upload responses

Dataset inspection responses can include metadata, row counts, processing state, column names, lineage, and a preview. Upload responses identify the created dataset; inspect it with get_dataset before treating the upload as ready.

​
Match and enrichment responses

Match responses can contain candidate identifiers, confidence, and a matched dataset or workflow run. Enrichment and contact-enrichment responses can return a workflow_run_id and an output dataset ID only after polling completes. Preserve the exact IDs and inspect the terminal result before starting the next operation.

​
Account and operation context

Account-scoped responses should retain account_id. A response may also carry session_id, search_handle, agent_run_id, workflow_run_id, dataset_id, output_dataset_id, a quote, or an error code. These values are opaque identifiers: do not derive one from another or replace one with a new request after a timeout.

​
Workflow response

Use get_workflow_run until the status is terminal. A failed or cancelled workflow must be inspected before deciding whether a new request is appropriate.

​
Confirmation and error response

See MCP limits and credits and MCP troubleshooting.