Skip to main content

Troubleshooting Cube connectivity

Connect docs via MCP

Why does test authentication fail?​

If the connection test fails when you provide your Cube credentials, check the following:

  • The Tenant URL must be the Cube host only, with no https:// prefix and no trailing slash or path—for example your-org.cubecloud.dev, not https://your-org.cubecloud.dev/.
  • The API Key must be a valid, unrevoked Cube API key with access to the deployment and environment you want to catalog.
  • The API key must be able to reach the deployment's control-plane API—confirm it hasn't been scoped away from the target deployment.

Why is my Cube catalog incomplete after crawl?​

  • Deployment or environment selection. The crawl extracts exactly one deployment and one environment. Confirm you selected the deployment and environment that hold the model you expect.
  • Security context. If you set a Security Context, Cube applies its security and row-level rules to that context, so the crawl sees only the model visible to that role or tenant. Clear the field to crawl the full model the API key can access.
  • API-key scope. The key governs which deployments and environments Atlan can read. A key scoped to a subset returns only that subset.

Why don't I see warehouse → Cube lineage in graph?​

Warehouse → Cube lineage is only emitted when Cube exposes the upstream database name for each data source and the matching Atlan warehouse connection exists. If lineage is missing:

  • Confirm the upstream warehouse connection (Snowflake, BigQuery, Redshift, Databricks, or another supported source) already exists in Atlan and has been crawled at least once, so its Table and Column assets are present when the Cube App runs.
  • Confirm Cube's data-source metadata exposes the database name for each connection—without it, Atlan can't build the warehouse identity needed to resolve lineage.
  • Review the crawl logs: the connector logs one lineage-diagnostics line per batch, including how many table and column references it saw and how many upstream Process entities it emitted.

What are known limitations of Cube connector?​

Atlan currently doesn't support the following:

  • Writing Atlan terms, classifications, or owners back to Cube—the connector is read-only ingest.
  • Cataloging more than one deployment-and-environment per workflow run—configure a separate workflow per deployment or environment you want cataloged.