Skip to main content
Native Databricks MCP setup in Sidecar is coming soon. Databricks managed MCP servers already use Streamable HTTP, so you can register one through Sidecar’s custom connector with a service-principal personal access token (PAT).
Databricks recommends OAuth for production. Sidecar’s custom connector does not yet perform OAuth token refresh, so this guide uses a service-principal PAT. Set a short lifetime that fits your operating process and rotate it before it expires.

​
Choose a managed server

Databricks publishes these endpoint patterns: See Databricks managed MCP servers for current endpoints, release stages, scopes, and resource permissions. Prefer a Genie Agent, AI Search index, or narrowly scoped Unity Catalog function when possible. The Databricks SQL server can modify data, and Sidecar applies MCP approval settings to the server as a whole rather than to individual tools.

​
Prepare the Databricks identity

  1. Create a Databricks service principal for Sidecar automation.
  2. Add it to the workspace.
  3. Grant it access to the selected MCP server and its underlying resource.
  4. For a SQL-backed resource, grant Can use on the SQL warehouse and only the required Unity Catalog privileges.
If you also connect Databricks as a Sidecar warehouse integration, follow the Databricks integration guide for the base service principal and Unity Catalog grants.

​
Create a service-principal PAT

A workspace admin can create a PAT on behalf of the service principal. Replace the application ID and choose a lifetime that matches your rotation policy. This example uses 90 days:
Copy the returned token value. See Databricks’ service-principal token management and create-obo-token CLI reference.

​
Register the server in Sidecar

Register each Databricks endpoint as a separate MCP server.
  1. In Sidecar, select Integrations → MCP Servers.
  2. Click Add MCP, then select Custom server.
  3. Enter values like these for a Genie Agent:
  1. Click Add server.
  2. Confirm that the server status is Connected and review the discovered tools.
The hostname must be your workspace hostname, not the account-console hostname or a SQL warehouse HTTP path. Databricks documents the bearer header and client setup in Connect MCPs to AI assistants and coding agents.
Custom MCP server form populated with a Databricks managed MCP endpoint

Register the exact managed-server endpoint as a custom MCP server.

Continue by granting the server to an agent and selecting it for jobs.

​
Operate the connection

  • Replace the token in Sidecar before the PAT expires.
  • Use one service principal per security boundary when different agents need different access.
  • Add Sidecar’s outbound addresses to Databricks workspace IP access lists when your workspace restricts clients.
  • Test the exact endpoint and PAT with the MCP Inspector when tool discovery fails.
Databricks-hosted custom MCP servers do not support PAT authentication. The direct PAT setup above applies to Databricks managed MCP servers and MCP Services. Put a compatible bearer-authenticated gateway in front of a custom Databricks-hosted server before registering it in Sidecar.