Model Context Protocol (MCP) on Databricks
This page is an overview of the MCP options on Databricks. MCP is an open source standard that connects AI agents to tools, resources, prompts, and other contextual information.
The main benefit of MCP is standardization. You can create a tool once and use it with any agent—whether it's one you've built or a third-party agent. Similarly, you can use tools developed by others, either from your team or from outside your organization.
MCP is one of three tool approaches available in Databricks Agent Framework. For guidance on when to choose MCP over Unity Catalog function tools or agent code tools, see Choose your tool approach.
Databricks MCP options
Databricks provides three MCP options:
- Managed MCP servers - Ready-to-use Databricks services
- External MCP servers - Third-party MCP servers, installable using Databricks Marketplace or custom Unity Catalog connections
- Custom MCP servers - Host your own MCP server as a Databricks app
Aspect | Databricks-managed MCP servers | External MCP servers | Custom MCP servers |
|---|---|---|---|
Intended use case | Databricks has ready-to-use servers that let agents query data and access tools in Unity Catalog. | Connect to third-party MCP servers hosted outside Databricks. Install popular servers from Databricks Marketplace or connect to any server using Unity Catalog connections. | Securely host your own MCP server as a Databricks app to bring your own server or run a third-party MCP server. |
Available tools | Expose specific Databricks services as MCP resources:
| Access external services and APIs:
| Bring your own custom tools and specialized business logic |
Setup complexity | Ready to use immediately | Install from Marketplace or create a custom Unity Catalog connection | Requires app deployment |
Security model | Unity Catalog permissions are always enforced, so agents and users can only access tools and data they're allowed to. | Databricks manages authentication using Unity Catalog connections with OAuth flows and token refresh. | You configure authentication and authorization. |
Authorization methods | Supports OAuth and PAT authentication to connect clients like Cursor and Claude Desktop. | Uses Unity Catalog connections for secure proxy authentication. | Only supports OAuth, which is unsupported by some clients like Cursor and Claude Desktop. |
To view your MCP servers, go to your workspace > Agents > MCP Servers:
Install servers from Databricks Marketplace
Databricks Marketplace offers a curated list of external MCP servers that you can quickly install to give you access to popular MCP servers hosted outside of Databricks. See Get access to external MCP servers
Compute pricing
Compute pricing for managed MCP servers depends on the MCP workloads:
-
Unity Catalog functions use serverless general compute. See serverless compute pricing.
-
Genie uses serverless SQL compute to run. See Serverless SQL pricing.
-
For the DBSQL server, SQL queries run on your SQL warehouses. See Databricks SQL pricing.
Custom MCP servers are subject to Databricks Apps pricing.