Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Connect Databricks Assistant to external tools and data sources through the Model Context Protocol (MCP). The Assistant can use any MCP servers that have been added to your workspace and that you have permission to use.
Note
MCP servers are only supported in Databricks Assistant agent mode.
What is MCP?
MCP is an open source standard that connects AI agents to tools, resources, prompts, and other contextual information. MCP provides a standardized way to expose tools, data, and workflows to the Assistant without embedding that context directly into prompts or instructions.
MCP is designed for cases where important context already exists but is difficult to access from an AI agent, like the Assistant. MCP offers a secure way to give Databricks Assistant access to context like the following:
- Documentation systems: Teams often keep runbooks or operational docs in tools like Confluence. Instead of copying sections into prompts, that content can be exposed once through an external MCP server and referenced when relevant. Your documents might also already be in Databricks as a vector search index, which the Assistant can use to query relevant documents.
- Tools and services: Platform teams may maintain APIs or scripts for setup, validation, or deployment. MCP allows those capabilities to be made directly available to the Assistant.
- Data sources: If you already have data in Databricks, you can connect the Assistant to a Genie space to help you analyze the data using natural language. For data stored outside of Databricks, you can give the Assistant secure access through external MCP servers.
In these cases, MCP replaces manual copy-pasting with a structured, reusable approach, making the right context available only when it is needed.
You can connect the Assistant to any MCP servers in your workspace that you have permission to use. Workspace admins control which servers are available, while users can select from those approved sources as needed.
To learn more about MCP servers in Databricks and how to create your own, see Model Context Protocol (MCP) on Databricks.
Add MCP servers to the Assistant
To connect the Assistant to an MCP server, add it in your Assistant settings:
- Open Assistant settings. In the Assistant panel, click
Settings. - Under MCP Servers, click
Add Server. - Select the MCP server(s) you want to add. You can select Databricks managed MCP servers or external MCP servers:
- Unity Catalog function: Select the schema(s) for the function(s). This allows the Assistant to use functions to run predefined SQL queries.
- Vector search: Select the schema(s) for the index(es). This allows the Assistant to query the vector search index to find relevant documents.
- Genie space: Select the Genie space(s). The Genie MCP invokes Genie as a tool, allowing you to query the Genie space(s) to analyze data using natural language.
- Unity Catalog connection: Select the connection(s) to use as external MCP servers. You must log in to the connection first before it can be used.
- Click Save.
After they've been added, MCP servers are immediately available to Databricks Assistant. You can start using them in agent mode without changing your prompts or instructions. The Assistant will automatically draw on these servers when relevant.
Currently, MCP access is limited to 20 tools across all servers. You can choose which tools and servers to enable or disable in your Assistant settings.