Databricks has a hosted MCP endpoint, Genie One, that Grok can reach as a custom connector. Contextflo is the other route, and it fits better when you want a broad set of tables and a shared layer of definitions for a team.
Both routes work from Grok because both are hosted. Neither needs anything running on your machine.
Option 1: Databricks' Genie One MCP in Grok
The endpoint lives on your workspace:
https://<workspace>/ai-gateway/mcp-services/system.ai.genie_one_mcp
It signs in with OAuth. The older beta endpoint is deprecated on October 31, 2026, so use this one.
Genie works over Genie spaces. A space can hold up to 50 tables, but Databricks recommends five or fewer for accuracy. So you pick a narrow slice of data per space and keep the questions inside it.
In Grok:
- Open grok.com/connectors.
- Choose New Connector, then Custom.
- Enter the endpoint URL with your workspace host.
- Complete the sign-in.
Option 2: Contextflo
Contextflo connects to a Databricks SQL warehouse.
On the Databricks side:
- In SQL Warehouses, open your warehouse and copy the HTTP path from Connection details.
- Create a token for a service principal or use a personal access token. Use an identity with read access only.
- In Unity Catalog, grant
USE CATALOG,USE SCHEMAandSELECTon what you want to expose.
In Contextflo: go to Data Sources, then Add data source, and pick Databricks. Enter the host (your workspace URL), the HTTP path and the token. Pick the tables and invite your team.
In Grok: add a custom connector at grok.com/connectors with:
https://mcp.contextflo.com/mcp
Each person signs in with their own account. Contextflo runs SELECT only.
Contextflo's access control is at the table level. Unity Catalog adds row filters and column masks, so keep those in place on the data side if you rely on them.
Which one to use
| Genie One MCP | Contextflo | |
|---|---|---|
| Scope | Genie spaces, five or fewer tables recommended | The tables you pick |
| Sign-in | OAuth with Databricks | Contextflo account per person |
| Shared context | Defined in the Genie space | Generated descriptions plus team corrections |
| Audit trail | Databricks' own query history and audit logs | Prompts and the SQL that ran, per person |
Just you, with a focused question set: option 1. Your team asks too, across more tables: Contextflo. For Claude, see connecting Claude to Databricks.
You do the modeling, we do the rest. Connect Databricks free.
FAQ
Does Databricks have a hosted MCP server Grok can reach? Yes. Databricks exposes a Genie One MCP endpoint on your workspace at https://<workspace>/ai-gateway/mcp-services/system.ai.genie_one_mcp, with OAuth. Because it is a hosted URL, a Grok custom connector can use it.
Is the older Databricks Genie MCP beta endpoint still supported? No, it is deprecated as of October 31, 2026. Use the Genie One MCP endpoint instead.
How many tables can a Genie space hold? Up to 50, but Databricks recommends five or fewer for accuracy. That limit shapes which route fits a broad set of questions.
What does Contextflo need from Databricks? The workspace host, the HTTP path of a SQL warehouse, and a token with read access. Grant USE CATALOG, USE SCHEMA and SELECT in Unity Catalog on what you expose.
See it on your own data.
Connect your warehouse and your team can ask Claude about it the same day, without waiting on a data team. Want a walkthrough first? Grab 20 minutes.
Book a 20-minute callSee how teams use Contextflo
Related posts
Keep reading

Conversational analytics: 5 ways to set it up, compared
7 min read

How to connect Claude to BigQuery? What works and what doesn't in 2026
8 min read

How to connect Claude to Postgres without giving it write access
9 min read

How to build a BI dashboard with Claude that your team can actually trust
8 min read

How to give Claude read-only access to your database (it isn't always the default)
7 min read

You connected your warehouse to Claude, now what?
5 min read

ClickHouse MCP: how to connect ClickHouse to Claude, Cursor or Codex
6 min read

Query BigQuery from the Codex CLI with MCP
5 min read

Run ClickHouse queries from Codex with MCP
6 min read


