Connect Databricks to AI: Let Claude and Cursor Query Notebooks, Tables, and Jobs

Why Connect Databricks to AI?

Databricks is where data teams run notebooks, manage tables, orchestrate jobs, and build ML pipelines. When you need to check a table schema, verify a job ran successfully, or read notebook output mid-coding, you open the Databricks workspace, navigate to the right resource, read it, switch back. For teams with hundreds of notebooks and dozens of scheduled jobs, that context switch adds up.

With an MCP server pointed at Databricks, AI agents can query your workspace directly. Check table schemas, read job statuses, browse notebook outputs without leaving your terminal.

What AI Agents Can Do With Databricks Access

Once connected, Claude, Cursor, or Codex can:

• Inspect tables — "What columns does the fact_orders table have?"
• Check jobs — "Did the nightly ETL job run successfully?"
• Browse notebooks — "What does the feature engineering notebook output?"
• Query catalogs — "What tables exist in the gold schema?"
• Read results — "What's the row count for the events table today?"
• Check clusters — "Is the analytics cluster running?"

How It Works

1. Open DataFaucet and enter your Databricks workspace URL
2. Browse Databricks — click through notebooks, tables, jobs, catalogs
3. DataFaucet captures the read operations as MCP tools
4. Deploy — your AI agents now have structured access to query Databricks

You're browsing the Databricks web workspace, not sharing personal access tokens. DataFaucet captures HTTP interactions from the web UI. Your Databricks credentials stay in your browser session.

Example Queries

"What tables exist in the gold.analytics schema?"
"What columns does the dim_customers table have?"
"Did the daily_pipeline job succeed this morning?"
"What's the last output of the revenue_forecast notebook?"
"How many rows are in the raw_events table?"

When This Saves Time

• Pipeline debugging — check job status and outputs without opening Databricks
• Feature development — validate table schemas before writing transforms
• Code review — verify a PR's assumptions about data structure
• Incident response — check failed jobs from terminal during outages
• Planning — query data catalog during architecture discussions

Security Model

DataFaucet connects through the Databricks web workspace. Access is scoped to whatever your Databricks permissions allow. If you only have access to certain workspaces or catalogs, the MCP server has the same constraints. No personal access tokens stored separately.

Getting Started

Open Databricks in your browser, copy the workspace URL. Paste it into DataFaucet, browse through notebooks, tables, and jobs. Deploy. Your agents can now query your lakehouse from your terminal.

Create your Databricks MCP server →