Skip to Help content
Back to Help

Upload fallback, waits, retries, and cancellation

Recover when a host cannot move a CSV, a wait returns early, or a run needs retry or cancel.

What if the host cannot upload my CSV?

MCP never reads a local filesystem path. If the host cannot move bytes, skip the transfer ticket.

  1. Open Datasets and upload the CSV in the browser.
  2. Return to the host and rediscover that Dataset. Stay at or under 25 MB.
  3. If a transfer ticket expired, refresh it through upload_csv with the same unfinished upload ID rather than starting a second Dataset.
  4. For dashboard upload failures that are not MCP transfers, use An upload failed.

How do I wait, retry, or cancel?

A wait is capped per call, not per Run. The Run continues if the host disconnects.

  1. Keep calling bounded get_run waits after you asked the host to wait. Do not prompt again only to continue waiting.
  2. Resume a paused Run after Usage and Billing show spendable value. Do not start a duplicate full Run over the same work unless you asked for that.
  3. Retry only when Everyn offers a retry for failed or unresolved rows. Do not blindly retry quality or spec failures.
  4. Cancel only when you asked to stop. Cancellation is harvest-aware; completed work stays inspectable.

Read Runs for sample versus full behavior, and A run is stuck, paused, or failed for dashboard recovery.

Was this article helpful?