How to use Databricks data from Cursor

Cursor can talk to Databricks through the hosted Genie One MCP endpoint, or through Contextflo connected to a SQL warehouse. Genie is built around curated spaces. Contextflo gives a whole team the same tables, definitions and an audit trail.

Option 1: Databricks Genie One MCP in Cursor

Databricks hosts an MCP endpoint at:

https://<workspace>/ai-gateway/mcp-services/system.ai.genie_one_mcp

It uses OAuth. Add it to ~/.cursor/mcp.json (global) or .cursor/mcp.json (a single repo):

{
  "mcpServers": {
    "databricks": {
      "url": "https://<workspace>/ai-gateway/mcp-services/system.ai.genie_one_mcp"
    }
  }
}

Cursor runs the OAuth flow when the server needs it. The older beta endpoint is deprecated on October 31, 2026, so use this one.

Genie works from spaces. A space holds up to 50 tables, and Databricks recommends five or fewer for accuracy. Plan one space per topic.

Option 2: Contextflo

In Databricks, create a service principal or use a personal access token, and grant it read access in Unity Catalog on what you want to expose:

• USE CATALOG
• USE SCHEMA
• SELECT

Then find the HTTP path: SQL Warehouses, your warehouse, Connection details.

In Contextflo, open Data Sources, then Add data source:

Pick tables, invite the team, then add the server to Cursor:

{
  "mcpServers": {
    "contextflo": {
      "url": "https://mcp.contextflo.com/mcp"
    }
  }
}

Each person signs in with OAuth. Contextflo runs SELECT only. Access is per person at the table level. Unity Catalog row filters and column masks still apply underneath if you use them.

Table and column descriptions are generated, corrections are saved as shared context, and every prompt and the SQL that ran lands in an audit trail. A weekly digest rounds it out.

Which one to use

Solo, one topic, a few tables? Genie One is a fine start. For a team asking across many tables from their editors, use Contextflo: shared context, per-person table access, and an audit trail of prompts and the SQL that ran.

You do the modeling, we do the rest. Connect Databricks free. Using Claude? See Claude with Databricks.

FAQ

What is the Genie One MCP endpoint for Databricks? It is Databricks' hosted MCP endpoint at https://<workspace>/ai-gateway/mcp-services/system.ai.genie_one_mcp, secured with OAuth. The older beta endpoint is deprecated on Oct 31, 2026.

How many tables can a Genie space hold? Up to 50. Databricks recommends five or fewer for accuracy, so a Genie space suits a narrow topic.

Which Databricks permissions does Contextflo need? A token with read access, from a service principal or a personal access token. In Unity Catalog, grant USE CATALOG, USE SCHEMA and SELECT on what you expose.

Does Contextflo honor Unity Catalog row filters? Contextflo controls access at the table level. Unity Catalog adds row filters and column masks on top, which is where you set those.