Install guide and config at curatedmcp.com

Databricks MCP: Give Claude Direct Access to Your Lakehouse

If you're running analytics or data engineering workflows on Databricks, you've probably wished your AI tools could actually do something with all that data instead of just talking about it. Databricks MCP closes that gap.

This is the official integration that lets Claude, Cursor, and other AI agents execute notebooks, query Delta tables, manage clusters, and inspect your data lineage—all from within your chat. Think of it as giving your AI a direct line to your lakehouse instead of just read-only documentation.

What It Does

Databricks MCP unlocks three core capabilities:

Data access. Query Delta Lake tables and Unity Catalog schemas directly. Your AI can write SQL, inspect table metadata, and understand your data structure without context-switching to the Databricks UI.

Automation. Execute notebooks and run SQL analytics on demand. Instead of manually kicking off jobs, you can ask Claude to run a transformation, check the results, and iterate—all in conversation.

Compute & ML ops. Manage cluster state, access MLflow experiments and model registries, and inspect job histories. Your AI becomes a useful teammate for routine ops tasks.

The server handles authentication via Databricks credentials and integrates seamlessly into existing MCP-compatible tools.

How to Install

Install via npm:

npx -y @databricks/mcp

Then add to your Claude Desktop config (typically ~/.config/Claude/claude_desktop_config.json on Linux/Mac):

{
  "mcpServers": {
    "databricks-mcp": {
      "command": "npx -y @databricks/mcp",
      "env": {
        "DATABRICKS_HOST": "your-workspace-url",
        "DATABRICKS_TOKEN": "your-pat-token"
      }
    }
  }
}

Store credentials securely—never commit tokens to version control. Refer to the official docs for Cursor and Windsurf setup.

Real-World Use Cases

• Debugging pipeline failures. "Run this notebook and show me what broke in the transformation step"—get results in seconds instead of manually checking the job UI.

• Ad-hoc analysis. "Query our customer table, count rows by region, and highlight anomalies"—your AI writes the SQL and surfaces insights without you touching the cluster.

• Model iteration. "Check the latest MLflow run, compare metrics to the last three experiments, and suggest which one to promote to staging"—keep experiments and context in one place.

Full install guides for Claude Desktop, Cursor, Windsurf, and more at CuratedMCP.