Set up the Databricks MCP server

Databricks ships managed MCP servers built into the platform — Unity Catalog functions, Genie spaces, SQL warehouses, and AI Search indexes — with no infrastructure to deploy. MintMCP exposes these as four recommended connectors (Databricks Functions, Genie, SQL, AI Search), and this guide covers finding your workspace URLs, registering a Databricks App Connection (or generating a PAT), and installing the connectors you need.

Prerequisites

• A MintMCP admin account
• A Databricks workspace with the Managed MCP Servers public preview enabled

Set up authentication

Every connector authenticates one of two ways — pick the one that fits, then paste the result into each connector you install. An OAuth App Connection keeps permissions per user and suits production; a PAT is quicker when a single shared service identity is good enough.

OAuth App Connection

An account admin registers one OAuth app in the Databricks account console. The same Client ID and Client Secret get pasted into each of the recommended Databricks connectors you install.

1. Go to the Databricks account console → Settings → App Connections.

2. Click Add connection.

3. Fill the form:

   Field	Value
   Name	A descriptive identifier (for example, mintmcp-databricks)
   Client type	Confidential (with secret)
   Scopes	genie, ai-search, sql, unity-catalog
   Redirect URL	https://app.mintmcp.com/oauth/callback
   Access Token TTL	1440 (24 hours)
   Refresh Token TTL	129600 (90 days)

   Note on scopes: all-apis still works as a scope but is broader than necessary. vector-search remains supported as the legacy alias for ai-search.

4. Click Save, then copy the Client ID and Client Secret. Databricks shows the Client Secret only once — copy it before you close the dialog.

Personal access token

A PAT is the simplest option when a single shared service identity is good enough — no account-admin involvement, no per-user OAuth.

1. In your Databricks workspace, click your username in the top bar and select Settings.
2. Click Developer.
3. Next to Access tokens, click Manage.
4. Click Generate new token.
5. Enter a description (for example, MintMCP gateway) and set a token lifetime. 90 days is a reasonable upper bound; plan to rotate before expiry.
6. Click Generate and copy the token — it's shown only once.

Note on PAT scope: A PAT inherits every permission the token owner has in the workspace — wider than any of the narrow OAuth scopes. Prefer OAuth for production.

Install a Databricks connector

In your Databricks workspace, go to Agents > MCP Servers to find your workspace URLs. Each MintMCP recommended connector wraps one Databricks managed server — pick the tab for the one you're setting up.

Databricks Functions

Runs Unity Catalog functions as MCP tools — a whole schema, or a single function.

Field	Value
URL	https://<workspace-hostname>/api/2.0/mcp/functions/{catalog}/{schema} (append /{function_name} for one function)
Scope	unity-catalog
Access	Read and execute, bounded by EXECUTE on the catalog, schema, or function

Add to MintMCP

1. In MCP store → Manage store → Recommended servers, search Databricks and Install the Databricks Functions connector.
2. In Server URL, paste the URL above.
3. Under Authorization Method, choose your auth and finish:

OAuth

• Select OAuth (detected from server).
• Open OAuth Client Advanced Settings and paste your App Connection's Client ID and Client Secret.
• Set Connection type to Per-user credentials.
• Click Install.

Bearer Token

• Select Bearer Token.
• Paste your PAT.
• Set Connection type to Shared credentials.
• Click Install.

Other endpoints: expose the system Python interpreter (https://<workspace-hostname>/api/2.0/mcp/functions/system/ai) or a single-function URL outside the catalog/schema pattern through + Add an MCP to your registry → Connect to an MCP by URL — the auth choices work the same way.

Security considerations

• Per-user EXECUTE with OAuth + per-user credentials — each user's Unity Catalog grants apply.
• Prefer OAuth — a PAT inherits every workspace permission its owner has.
• Public preview — no production SLA yet; allowlist MintMCP's outbound IPs if the workspace restricts inbound. Unity AI Gateway (April 2026) adds Databricks-side governance to pair with MintMCP gateway logs.

Databricks Genie

Queries a single Genie space in natural language.

Field	Value
URL	https://<workspace-hostname>/api/2.0/mcp/genie/{genie_space_id}
Scope	genie
Access	Read-only, bounded by the Genie space's underlying dataset permissions

Add to MintMCP

1. In MCP store → Manage store → Recommended servers, search Databricks and Install the Databricks Genie connector.
2. In Server URL, paste the URL above.
3. Under Authorization Method, choose your auth and finish:

OAuth

• Select OAuth (detected from server).
• Open OAuth Client Advanced Settings and paste your App Connection's Client ID and Client Secret.
• Set Connection type to Per-user credentials.
• Click Install.

Bearer Token

• Select Bearer Token.
• Paste your PAT.
• Set Connection type to Shared credentials.
• Click Install.

All spaces at once: front every Genie space the user can access through the workspace-aggregated Genie One endpoint https://<workspace-hostname>/api/2.0/mcp/genie (no space ID) via + Add an MCP to your registry → Connect to an MCP by URL — same auth choices.

Security considerations

• Read-only — no write path through the Genie MCP server.
• Dataset permissions apply — Genie returns only rows the connected identity can see.
• Prefer OAuth — a PAT inherits every workspace permission its owner has.
• Public preview — no production SLA yet; allowlist MintMCP's outbound IPs if the workspace restricts inbound. Unity AI Gateway (April 2026) adds Databricks-side governance to pair with MintMCP gateway logs.

Databricks SQL

Runs SQL against Databricks SQL warehouses.

Field	Value
URL	https://<workspace-hostname>/api/2.0/mcp/sql
Scope	sql
Access	Read and write SQL, bounded by warehouse access and Unity Catalog table permissions

Add to MintMCP

1. In MCP store → Manage store → Recommended servers, search Databricks and Install the Databricks SQL connector.
2. In Server URL, paste the URL above.
3. Under Authorization Method, choose your auth and finish:

OAuth

• Select OAuth (detected from server).
• Open OAuth Client Advanced Settings and paste your App Connection's Client ID and Client Secret.
• Set Connection type to Per-user credentials.
• Click Install.

Bearer Token

• Select Bearer Token.
• Paste your PAT.
• Set Connection type to Shared credentials.
• Click Install.

Security considerations

• Supports writes — agents can run mutating SQL; restrict write tools via Tool customization if that's a concern.
• Warehouse + table permissions apply — only reachable tables are readable or writable.
• Prefer OAuth — a PAT inherits every workspace permission its owner has.
• Public preview — no production SLA yet; allowlist MintMCP's outbound IPs if the workspace restricts inbound. Unity AI Gateway (April 2026) adds Databricks-side governance to pair with MintMCP gateway logs.

Databricks AI Search

Runs semantic search against an AI Search index.

Field	Value
URL	https://<workspace-hostname>/api/2.0/mcp/ai-search/{catalog}/{schema}/{index_name}
Scope	ai-search
Access	Semantic-search read, bounded by index access permissions

Renamed service: Databricks renamed Vector Search to AI Search. The legacy /api/2.0/mcp/vector-search/ URL prefix and vector-search scope still work for backward compatibility.

Add to MintMCP

1. In MCP store → Manage store → Recommended servers, search Databricks and Install the Databricks AI Search connector.
2. In Server URL, paste the URL above.
3. Under Authorization Method, choose your auth and finish:

OAuth

• Select OAuth (detected from server).
• Open OAuth Client Advanced Settings and paste your App Connection's Client ID and Client Secret.
• Set Connection type to Per-user credentials.
• Click Install.

Bearer Token

• Select Bearer Token.
• Paste your PAT.
• Set Connection type to Shared credentials.
• Click Install.

Security considerations

• Read-only — semantic search only, no write path.
• Index permissions apply — the connected identity needs read on the AI Search index.
• Prefer OAuth — a PAT inherits every workspace permission its owner has.
• Public preview — no production SLA yet; allowlist MintMCP's outbound IPs if the workspace restricts inbound. Unity AI Gateway (April 2026) adds Databricks-side governance to pair with MintMCP gateway logs.

Troubleshooting

• redirect_uri_mismatch on OAuth install — the App Connection's redirect URL must be exactly https://app.mintmcp.com/oauth/callback.
• 401 Unauthorized with a PAT — the token expired or was revoked, so regenerate it in Databricks and update the connector.
• Empty tool list after install — the connected identity lacks the permission the connector needs: EXECUTE on the catalog or schema (Functions), Genie space membership and dataset permissions (Genie), warehouse and table access (SQL), or read on the index (AI Search).
• vector-search scope rejected — the account admin disabled legacy scopes, so switch the App Connection to ai-search.

Next steps

• Tool customization — Control which Databricks tools are exposed to users
• MCP gateway administration — Manage access and permissions
• Unity AI Gateway docs — Databricks-side governance for MCP traffic