Databases & CRM
Set up Databricks SQL
Connect a Databricks Unity Catalog schema through a SQL warehouse. Tables become searchable entries, and tables with declared keys become records with their figures totalled exactly — over the Statement Execution API with a personal access token, no driver and no app registration. Read-only.
Before you start
- A personal access token — your avatar → Settings → Developer → Access tokens → Generate new token — best issued for a user whose grants are read-only on the catalog; only reads are ever issued regardless.
- A running SQL warehouse (SQL Warehouses in Databricks) — Serverless starts fastest; a stopped classic warehouse can take a couple of minutes to wake on the first sync of the day.
- Unity Catalog: the legacy hive_metastore can't be introspected this way, so your tables need to be in Unity Catalog.
- Documents → Databricks SQL: enter Workspace URL, SQL warehouse (its ID from Connection details — pasting the whole HTTP path also works), and Catalog, and paste the token.
- Schema defaults to default; set it only if your tables live elsewhere.
- Optionally narrow with Only these tables or Skip these tables.
- Choose access groups, Test, then Sync.
What comes in
- Every table as a searchable entry — columns, types, sample values, and what it links to.
- Tables with declared (informational) primary keys also become one record per row, with parent names and an exact summary of child facts; Unity Catalog constraints are often left undeclared on lakehouse tables, so a table without one still shows up, just without records.
| Schema | The schema inside the catalog to read — blank uses default. |
|---|---|
| Only these tables | Allow-list, one table per line; blank reads the whole schema. |
| Skip these tables | The opposite — read everything except these. |
Access
Visible only to the access groups chosen on the source — Databricks' own catalog grants aren't read row by row; the token's own grants only cap what the connector can reach at all.
A serverless warehouse answers fastest; a stopped classic one can take a couple of minutes to wake on the first sync of the day. Records are capped at the newest 5,000 rows per table, the same as every SQL-shaped source in this section.
Was this page helpful?
Last updated 20 Sep 2026