Skip to main content
Applies to:
  • Plan - Any
  • Deployment - Any
  • Use case - Resolve dataset lookup errors despite browser access

Summary

MCP resolve_object can return Missing read access when a dataset URL or name lookup finds no accessible match. If you can open the dataset in the browser, retry with its UUID to bypass name lookup.

What is happening

Dataset browser URLs contain a URL slug. When you pass one to MCP resolve_object, it uses the slug as object_name and searches the dataset’s display name. If the slug differs from the name, the lookup can fail before any dataset rows are queried. An empty lookup result can reflect a name mismatch or missing read permissions. The error alone does not distinguish these causes. UUID resolution bypasses name lookup but still checks read access.

Fix or suggestion

  1. List datasets with your personal API key, filtering by org_name and project_name. Leave dataset_name unset so a name mismatch does not hide the dataset. Replace the placeholders with your organization name, project name, and API key:
    Use your organization’s Braintrust API URL in place of https://api.braintrust.dev if you connect to another MCP endpoint, such as EU or BYOC. Find the dataset in the response’s objects array and copy its id and name. If needed, paginate with starting_after set to the last object’s id from the previous page.
  2. Call MCP resolve_object with only the dataset type and UUID. Omit url, object_name, and the other name fields:
  3. Use the returned object_id for subsequent dataset queries.

Option 2: Disambiguate name-based lookups

  1. Set object_name to the exact name returned by the List datasets API, rather than its url_slug.
  2. Include both org_name and project_name to scope the lookup. Omit url and object_id:

How to confirm it worked

  • The tool result’s text content contains JSON with the expected object_id, object_name, and permalink, without an error. Success does not require an explicit isError: false field.
  • Dataset row queries using that object_id succeed with the same credentials.

Notes

If UUID resolution also fails, retrieve the dataset by ID through the REST API using the same personal API key used for MCP. Compare the HTTP status and error response, and verify that the MCP account has read permissions for the dataset. Share the UUID, exact MCP arguments, and error responses with support, omitting credentials.