Use Databricks with AI via MCP
The lakehouse holds the numbers people need most often – locked behind SQL and notebooks. Databricks provides official managed MCP servers that run serverless inside your own workspace. CompanyGPT connects to them directly, and the permissions granted in Unity Catalog continue to apply unchanged.
What is Databricks?
Databricks is a data platform for analytics and AI that combines data warehouse and data lake into a single architecture, known as the lakehouse. Organizations process large volumes of data there, build analyses and models, and manage access rights, lineage and metadata centrally through Unity Catalog. The platform runs in the cloud, including EU regions on AWS, Azure and Google Cloud.
For AI agents, Databricks matters because it holds the properly modeled company metrics – not the copies that end up in spreadsheets and slide decks. Once an agent can reach that source, it answers questions from the same data reporting and controlling rely on. What makes this workable is that the permission model travels with the request instead of being bypassed by a shared service account.
Connection at a Glance
- MCP server
- Official (Databricks), managed in your own workspace, public preview
- Endpoint
- https://<workspace>/api/2.0/mcp/…
- Transport
- Streamable HTTP
- Authentication
- OAuth (Databricks identity)
- Operating model
- SaaS (customer cloud workspace)
- Permissions
- Unity Catalog permissions are inherited
- Covered services
- Genie One, Genie Agent, AI Search (formerly Vector Search), Databricks SQL, Unity Catalog functions
How to Connect Databricks to CompanyGPT
Databricks operates official managed MCP servers that run serverless inside the customer’s workspace. Per the vendor documentation (as of 17 August 2026) they are in public preview. They cover Genie One and Genie Agent for natural-language questions, AI Search (formerly Vector Search), Databricks SQL, and functions registered in Unity Catalog. The endpoints sit under https://<workspace>/api/2.0/mcp/… – no additional third-party infrastructure is introduced, and the endpoint belongs to your own Databricks workspace.
Authentication runs through OAuth against Databricks. The point that matters most for governance owners: Unity Catalog permissions are inherited. An agent sees exactly the catalogs, schemas and tables the connected identity is entitled to – no second permission model appears alongside Databricks.
On the CompanyGPT side, your administrators decide which roles and groups can see the Databricks integration at all and which agents it is attached to. Access can be limited to controlling and department leads, for example, while all other teams continue to work without data access.
Typical Use Cases
How AI agents in CompanyGPT work with Databricks
Ask for metrics in plain language
A department lead asks about revenue development or capacity utilization in chat and gets the answer from a Genie space – no SQL, no detour through the BI team.
Make data products discoverable
New members of the data team can have it explained which tables and functions exist in Unity Catalog for a given topic and what they are intended for.
Connect analysis and document
Results from the lakehouse flow straight into a management report or decision paper, combined with other CompanyGPT integrations.
Frequently Asked Questions
Databricks and AI – the most important answers
Does Databricks have an official MCP server?
Yes. Databricks provides managed MCP servers, in public preview per the vendor documentation (as of 17 August 2026). They run serverless inside your workspace and cover Genie One, Genie Agent, AI Search (formerly Vector Search), Databricks SQL and Unity Catalog functions.
Does the AI get access to all data in the lakehouse?
No. Unity Catalog permissions apply unchanged. An agent can only reach the objects the connected Databricks identity is entitled to.
What does our data team need to prepare?
MCP access is enabled in the workspace and the relevant Genie spaces or functions are released. The endpoint is then registered in CompanyGPT and assigned by role or group.
Does our data leave our own environment?
CompanyGPT runs in your own cloud tenant and talks directly to the MCP endpoint of your Databricks workspace. No further service provider sits in between, and your choice of Databricks region still determines where the data resides.
Can the AI modify data in Databricks?
The possible scope follows from the released objects and the rights of the connected identity. For purely analytical scenarios a read-only identity is used, so write operations are technically ruled out.
Databricks and Your AI – GDPR-Compliant with CompanyGPT
CompanyGPT is the enterprise AI assistant in your own cloud tenant: all relevant AI models, your company data and integrations like Databricks – centrally managed, with roles and permissions. We set up the connection together with you.
Databricks is a trademark of its respective owner (Databricks). Its mention describes compatibility and does not imply any partnership or endorsement. Availability and scope of the connection depend on the respective vendor.