How to use Claude with Databricks

Databricks is increasingly where companies land when they need a unified platform for data engineering, ML, and analytics. With Unity Catalog and lakehouse architecture, it's become the warehouse of choice for data-heavy teams. Connecting it to Claude is straightforward, Databricks has official MCP support. But as with every warehouse, the connection is the easy part. Making it work for a team is where it gets interesting.

Option 1: Databricks MCP (direct connection)

Databricks offers managed MCP servers that connect Claude to your Unity Catalog data. You authenticate via OAuth, and Claude gets access to tables, vector indexes, and functions registered in Unity Catalog.

// claude_desktop_config.json
{
  "mcpServers": {
    "databricks": {
      "url": "https://<workspace>.databricks.com/api/2.0/mcp/server",
      "headers": {
        "Authorization": "Bearer <DATABRICKS_TOKEN>"
      }
    }
  }
}

This gives Claude access to query your lakehouse, explore schemas, and call Unity Catalog functions. For a data engineer or data scientist, it's a solid setup.

Limitations

• Technical users only: Databricks MCP gives Claude raw access to your catalog. Useful if you know the schema. Not useful if you're a product manager who doesn't know which table has conversion data.
• No business context: Unity Catalog has metadata (descriptions, tags, lineage) but Claude doesn't automatically know your business definitions. "Revenue" still needs to be defined somewhere.
• Per-person setup: Each team member configures their own OAuth connection. No centralized management.
• Access control is Databricks-native: You get Unity Catalog's permission model, which is powerful but complex to configure per person for an analytics use case.

Option 2: Databricks AI/BI Genie

Databricks has its own conversational analytics product, AI/BI Genie. It's built into the Databricks workspace, understands Unity Catalog metadata, and lets users ask questions in natural language.

It's a good option if your entire team already works inside Databricks. The tradeoff is vendor lock-in: your context, your definitions, your saved questions all live inside the Databricks ecosystem. If you ever want to use a different LLM or switch warehouses, you're rebuilding everything.

Option 3: Databricks + Claude via Contextflo

Contextflo connects to your Databricks workspace using a service principal or personal access token with read-only permissions. It scans your Unity Catalog schemas, auto-generates context, and serves it to Claude through MCP.

Your team asks questions in Claude, the tool they're already using for everything else. Claude uses the context layer to write accurate SQL against your Databricks lakehouse. Business definitions, metric calculations, and access controls are centralized and consistent.

Comparison

Which to choose

Databricks MCP is great for a data engineer or data scientist who wants to query their lakehouse from Claude without context-switching. Fast, free, technical.

AI/BI Genie makes sense if your whole company lives in the Databricks workspace and you're fine with the lock-in.

Contextflo is for teams where non-technical people need answers from your lakehouse through the AI tool they already use, with shared context and access controls that don't require configuring Unity Catalog permissions for every business user.