Databricks MCP: connect Databricks to Claude and ChatGPT
Ask Claude or ChatGPT what revenue per region was last quarter, how many customers signed up each week or which table holds the invoice data. The AI finds the right tables in Databricks, writes the query, runs it and answers in plain words, so people get numbers without writing SQL or waiting in the data team’s queue.
With Deployed OS everyone connects their own Databricks user. The AI sees the catalogs, tables and SQL warehouses that person may use, and it only reads: nothing in Databricks can be written, updated or deleted.
7 read actions · read-only · Works in Claude, ChatGPT and Copilot
What your team can do with Databricks in Claude and ChatGPT
From question to result faster
Numbers without a ticket to the data team
Follow usage and pipeline yourself
Governed access for the whole company
Ask it like this
“Which tables do we have in the sales schema of the main catalog, and what does each one contain?”
“What columns are in main.finance.invoices, and what does each one mean?”
“What was net revenue per region last quarter? Use main.sales.orders.”
“How many new customers signed up each week since September? Use the customers table in the crm schema.”
“Show the 20 accounts whose usage dropped the most this month compared with last month.”
“Which SQL warehouses can I use, and which one is running right now?”
What the Databricks MCP connector does today
- Reads
databricks_list_catalogsList the Unity Catalog catalogs you can see in your company’s Databricks workspace, with name, type, comment, owner and link.
- Reads
databricks_list_schemasList the schemas you can see in one catalog, with name, comment, owner and link.
- Reads
databricks_list_tablesList the tables and views you can see in one schema, with name, type, format, comment, owner and link.
- Reads
databricks_get_tableRead one table or view: its columns with type and comment, and its type, format, owner, comment and last change.
- Reads
databricks_list_warehousesList the SQL warehouses you can use, with name, state, size and whether each is serverless.
- Reads
databricks_run_queryRun one read-only SQL statement (SELECT, WITH, SHOW, DESCRIBE, EXPLAIN or VALUES) on a SQL warehouse as you, and get the columns and rows back, up to 1,000 rows.
- Reads
databricks_get_query_resultFetch the result of a query that was still running, or tell you that it has not finished yet.
Need it to do more? We extend connectors for the companies that use them.
Who sees what
- Read-only
- The connector browses catalogs, schemas, tables and views, and runs one read-only SQL statement at a time, up to 1,000 rows. Only SELECT, WITH, SHOW, DESCRIBE, EXPLAIN and VALUES are allowed. Nothing can be written, updated or deleted.
- Each person’s own Databricks access
- Everyone connects their own user, and queries run as that person on a SQL warehouse they may use. A table that finance cannot see in Databricks stays hidden from finance here too.
- Your company decides who gets it
- Admins choose which groups get Databricks in Deployed OS. Start with the analysts and add finance or sales ops when you are ready.
- Every call logged
- Who asked, which table was read or which query was run, when. Deployed OS shows the calls per connector and per person, and the log can be exported.
Set up the Databricks connector in three steps
Databricks MCP: common questions
Yes. Databricks offers its own managed MCP servers, switched on in your Databricks workspace. Deployed OS gives your whole company one connection with Databricks next to your other tools, with access per group and a log of every call.
Your admin switches on Databricks in Deployed OS once. You add your company’s Deployed OS connection in Claude, connect your own Databricks user and start asking. The same connection works in ChatGPT and Copilot.
No. It browses the catalog and runs read-only SQL, one statement at a time. Nothing can be written, updated or deleted, and statements that would change data are refused. If you want it to do more, we can extend it for you.
Yes. Each person works as their own Databricks user, so the AI sees the catalogs, tables and SQL warehouses that person may use and nothing else. The permissions you have set in Unity Catalog keep applying.
Deployed is an EU company and fully GDPR compliant, and you choose where your data is stored, down to the city. Table details and query results are fetched from Databricks when someone asks for them.
It depends on the size of your team. Get in touch and we will walk you through it.
See it on your own tools.
Tell us which systems your company runs on. We’ll show you what your team can ask, and what IT gets to control.