Databricks MCP
Query Unity Catalog, run notebooks and inspect jobs from an AI agent.
Last verified: September 20, 2026
How to Install Databricks MCP
$npx -y @databricks/mcpRequires Claude Desktop, Cursor, Windsurf, or another MCP-compatible client.
About Databricks MCP
The Databricks MCP server exposes the lakehouse to an AI agent: browsing Unity Catalog schemas, running SQL against warehouses, triggering notebooks and inspecting job runs. Because Unity Catalog governs access, the agent inherits existing permissions rather than bypassing them, which is the main reason data teams are willing to wire an assistant into production data at all. It is most useful for exploratory analysis and pipeline debugging rather than unattended writes.
Databricks MCP is a Database MCP server designed for developers, analysts, and data teams. It enables querying and managing databases from an AI assistant by exposing schema introspection and SQL execution as MCP tools. Key capabilities include schema introspection across tables and views, parameterised sql execution, read-only and read/write modes, and structured tool calls that work cleanly inside an LLM context. It integrates with Postgres, MySQL, SQLite, MongoDB, Redis, BigQuery, and Snowflake, and is best suited for developers, analysts, and data teams who need ad-hoc analytics without a bi tool.
Key Features
- lakehouse
- unity-catalog
- sql
- data-engineering
- notebooks
Pricing
- Core MCP server
- Community support
- Works with any MCP client
- Everything in Free
- Higher usage limits
- Priority support
- Everything in Pro
- SLA & SSO
- Dedicated support
Tier details are indicative — visit the Databricks MCP website for current pricing.
Pros & Cons
Pros
- Lets the assistant write and run real queries instead of guessing
- Read-only mode is safe to point at production-shaped data
- Skips most ORM and CSV-shuffling boilerplate
Cons
- Misconfigured permissions can expose sensitive data
- Long-running queries need explicit timeouts
- Schema drift requires reconnecting or refreshing context
Best For
- Ad-hoc analytics without a BI tool
- Drafting migrations and reviewing query plans
- Letting non-SQL teammates ask questions of a database
- Bootstrapping internal admin workflows
Screenshots
Screenshots coming soon —
Submit yours →Our Take on Databricks MCP
The right server if your data already lives in Databricks. Unity Catalog governance is what makes it safe to point an agent at; without a Databricks workspace it is irrelevant.