Databricks

databricks-sql-connector-backed connector for Databricks SQL Warehouses and Unity Catalog. db_type is databricks. Tier 2.

Connection config

databricks

{
  "name": "prod-databricks",
  "db_type": "databricks",
  "host": "adb-1234567890.azuredatabricks.net",
  "http_path": "/sql/1.0/warehouses/abcdef123456",
  "access_token": "...",
  "catalog": "main",
  "schema_name": "default"
}

Connection fields

Field	Required	Description
name	Yes	Connection name. [a-zA-Z0-9_-], max 64 chars.
db_type	Yes	databricks.
host	Yes	Workspace hostname (e.g. adb-….azuredatabricks.net), no scheme.
http_path	Yes	SQL Warehouse path (e.g. /sql/1.0/warehouses/abcdef123456).
access_token	One of	Personal Access Token (PAT) — default auth method.
catalog	No	Unity Catalog name.
schema_name	No	Default schema (mapped to schema).
query_timeout	No	Statement timeout, seconds (best-effort SET statement_timeout).

Auth methods: PAT (default, requires access_token); OAuth M2M (service principal — auth_method: oauth_m2m with oauth_client_id / oauth_client_secret); OAuth U2M (browser — auth_method: oauth_u2m). OAuth methods require databricks-sdk.

Capabilities

Capability	Supported	Notes
Query	Yes	Per-statement timeout via SET statement_timeout (best-effort; not all runtimes support it).
Schema introspection	Yes	information_schema.columns / .tables (Unity Catalog); falls back to SHOW SCHEMAS + SHOW TABLES + DESCRIBE TABLE for legacy Hive metastore.
FK discovery	Yes	information_schema.table_constraints joined to key_column_usage / constraint_column_usage (Unity Catalog only; best-effort).
Primary keys	Yes	table_constraints + constraint_column_usage (Unity Catalog).
EXPLAIN	Yes	EXPLAIN FORMATTED; row estimate parsed from the physical plan.
Cost estimation	Yes (estimated)	EXPLAIN FORMATTED row estimate × per-row rate; defaults to 10,000 rows when the plan has no rowCount/numOutputRows.
Schema stats	Partial	DESCRIBE DETAIL per Delta table (capped at 50 tables) for numFiles and sizeInBytes; no row counts.

Tier 2.

Dialect notes / gotchas

• Identifiers are quoted with backticks (`col`).
• FK and PK discovery only work on Unity Catalog; Delta tables do not enforce constraints, so they may be absent.
• Row counts are not collected — DESCRIBE DETAIL provides file count and byte size, not row count.
• Row-count batching and DESCRIBE DETAIL are limited to the first 50 tables.

Blocked functions

No Databricks-specific dangerous-function denylist is defined. Only the universal block applies:

• load_extension
• install_extension

All DDL/DML statement types are blocked (CREATE, DROP, ALTER, INSERT, UPDATE, DELETE, TRUNCATE, MERGE, GRANT, REVOKE, etc.).

Recommended plugin skill

sql-workflow — no dedicated Databricks skill exists. Covers output-shape inference, schema exploration, CTE-based query building, and the verification loop.

Cloud vs local

Supported in both cloud and local mode. In cloud mode the host is validated against the Databricks domain allow-list: *.cloud.databricks.com, *.azuredatabricks.net, *.gcp.databricks.com, *.databricksapps.com. Databricks connects over HTTPS, so it skips the TCP SSRF DNS check.