DATABASE QUERY AND ANALYSIS / Databricks
Databricks SQL MCP
Run explicit SQL against Unity Catalog tables through the managed system.ai.dbsql service, with caller permissions and optional write restriction.
“Databricks SQL MCP server”
01 / THE REASONING
Why this made the selection.
- Connects SQL execution to Unity Catalog and warehouse permissions rather than distributing a database credential to a local adapter.
- Documents asynchronous execution, result truncation and a policy switch for read-only operation.
- Databricks’ SQL page says this MCP server runs AI-generated SQL against Unity Catalog tables and permits reads and writes by default.
- The built-in services page says OAuth clients must request the listed scope and points readers at legacy workspace endpoints separately from the current servers.
02 / THE REVIEW RECORD
What we actually inspected.
Source review has boundaries.
A clear record is more useful than a “safe” badge.
Material inspected
- Databricks SQL MCP setup, policy and query limitations
- Databricks-provided MCPs and legacy endpoint boundaries
- https://docs.databricks.com/aws/en/agents/mcp-tools/databricks-sql
- https://docs.databricks.com/aws/en/agents/mcp-tools/built-in-mcp-services
- Round 2 reread of acquired publisher setup and permission evidence; see immutable wave review.
Our findings
- Current official documentation distinguishes the system.ai.dbsql Unity Gateway endpoint and ai-gateway scope from the legacy sql-scoped workspace endpoint.
- The SQL page explicitly documents write-default behaviour, read-only policy, asynchronous queries and result truncation.
- The SQL MCP page says the server runs AI-generated SQL against Unity Catalog tables from AI coding tools.
- That server permits reads and writes by default; system.ai.dbsql_policy supports disallow_writes=true for read-only operation.
- The built-in services page says to use the full server URL and, with OAuth, to request the listed scope. It also distinguishes current Databricks-provided MCPs from legacy workspace endpoints.
- The Unity Gateway SQL endpoint uses the ai-gateway OAuth scope and permits reads and writes by default; queries are asynchronous.
Not established by this review
- Installation, authentication and live tool execution were not tested.
- Service availability, security isolation and package/container parity with source were not tested.
- The resource was not executed and its tools were not called.
- The server was not called. A source review is not a runtime or security certification.
The review applies to the material and revision named here. A newer upstream release can change its behavior.
03 / PUT IT TO WORK
Connect a server deliberately.
- Have an account admin enable Unity Gateway beta and grant the needed MCP, Unity Catalog and SQL warehouse access.
- Connect the workspace URL https://<workspace-hostname>/ai-gateway/mcp-services/system.ai.dbsql using the ai-gateway OAuth scope and the official client setup guide.
- For read-only work, set disallow_writes=true in system.ai.dbsql_policy before connecting the agent, alongside appropriate database grants.
- Start with the documented SELECT 1 AS result check, verify the actual tool call and result, and handle polling. In programmatic clients, preset warehouse_id in _meta when warehouse selection must be deterministic.
- Have an account administrator enable the Unity Gateway beta, select the SQL warehouse, and set disallow_writes=true for read-only access.
Before you start
- A Databricks workspace with Unity Gateway beta enabled and a usable SQL warehouse.
- An authenticated caller with the required Unity Catalog, warehouse and MCP permissions.
04 / FOLLOW THE EVIDENCE
The source trail.
Our notes are separate from the original resource.
Check upstream before adopting a new version.
https://docs.databricks.com/aws/en/agents/mcp-tools/databricks-sql
Supports: install, access, limitations, bestFor, review
https://docs.databricks.com/aws/en/agents/mcp-tools/built-in-mcp-services
Supports: install, limitations, access