Databricks Genie Explained: Genie One, Agents, Code and MCP

Databricks Genie is now a family of tools rather than a single chat box. This guide explains what each part does, what changed in September 2026, how to connect Claude or ChatGPT to it, what it costs, and how accurate it was in my own test.

What is Databricks Genie?

Genie is Databricks’ family of natural-language tools for data. Genie One lets business users query data in plain English, Genie Agents are curated environments that hold the trusted data and business rules behind those answers, and Genie Code helps developers write and run data work inside the workspace.

What is a Genie Agent, and is it the same as a Genie space?

A Genie Agent is the curated environment a data team builds for a domain. It holds the datasets, instructions, sample queries, metrics and business rules that Genie One uses to answer questions, and every answer is governed through Unity Catalog. Earlier material called the closest equivalent a Genie space.

Databricks documentation confirms the rename: Genie Agents were previously called Genie spaces.

For a tested setup with six tables and fourteen benchmark questions, see my Databricks Genie tutorial on business context.

What changed in Databricks Genie in September 2026?

How do you connect Claude or ChatGPT to Genie?

The Genie One MCP server, generally available since September 25, 2026, lets MCP clients such as Claude, ChatGPT or Cursor query Genie. Unity Catalog permissions apply to each request. The earlier Beta endpoint is deprecated and will be sunset on October 31, 2026, so anything still using it needs to move.

What does Genie cost?

From the Databricks Genie page, checked October 26, 2026: Genie One and Genie Agents usage by users is free through January 31, 2027; Genie Code moved to pay-as-you-go with a monthly per-user free allowance from July 8, 2026; service principals are billed. Pricing changes, so check the current page before you budget.

How accurate is Genie on business questions?

In my test, Databricks scored the benchmark at 0% with neutral table information and 79% (eleven of fourteen) after I added business definitions; my manual review found thirteen of fourteen answers business-correct, or 93%. One generated SQL query still failed, returning 56.89% instead of the verified 4.11% return rate. The full method, prompts and failure are in the Databricks Genie tutorial.

Measuring AI answers against a baseline is the habit behind my book AI to ROI for Business Leaders.

Related reading

Official resources

Leave a Comment

Your email address will not be published. Required fields are marked *