databricks

This MCP server exposes a Databricks interface via the MCP (Model Context Protocol) so you can programmatically query and operate on Databricks workspaces. It provides tools to list catalogs, schemas, tables, and SQL warehouses, and to execute SQL statements against a Databricks SQL warehouse. Use the tools through MCP-compatible clients or IDE integrations to send commands like list_catalogs, list_schemas, list_tables, execute_sql, and list_warehouses, and receive structured JSON results that reflect the Databricks workspace state. Authentication is handled through Databricks unified authentication, so ensure your environment is configured with the appropriate credentials or token prior to making requests.

To interact with the server, connect your MCP client to the server’s standard MCP endpoints (stdin/stdout in the typical MCP pattern) and issue the desired tool calls. For example, you can request the list of catalogs to discover available Databricks catalogs, then drill down to schemas and tables within a chosen catalog/schema. When executing SQL, provide the SQL statement and optional timeout and row limit to control execution and result size. The server also lists available SQL warehouses so you can target the correct warehouse for your queries.

Prerequisites:

• A supported host (Linux, Windows, or macOS)
• Go 1.24+ if building from source (the README indicates Go-based server)
• Access to the Databricks workspace with appropriate permissions

Installation options:

1. Download the latest release binary from the project releases page and run it directly:

   • Locate the latest release for your platform
   • Unpack if needed
   • Run the server binary (e.g., ./databricks-mcp-server or databricks-mcp-server.exe)

2. Use the MCP client integration (via npx) in your development environment:

   • This is shown in the VS Code extension setup: npm-based install: npx -y databricks-mcp-server@latest
   • You can also wire the server into your MCP client using the standard MCP protocol (stdin/stdout).

Prerequisites recap:

• Ensure you have Node.js/npm if you plan to use the npx-based startup in an editor integration
• Ensure Go is installed if you intend to build from source or follow repository-specific build steps
• Ensure you have network access to your Databricks workspace and a valid authentication method configured

Basic run example (from the release binary):

• chmod +x databricks-mcp-server
• ./databricks-mcp-server

If you prefer the npx startup (as shown for editor integration):

• npx -y databricks-mcp-server@latest

Tips and notes:

• The server uses Databricks unified authentication; configure credentials/tokens per your Databricks setup before issuing tool commands.
• Supported platforms include Linux (amd64), Windows (amd64), macOS (Intel/amd64, Apple Silicon arm64).
• If you encounter connectivity or authentication errors, verify that your Databricks workspace URL and token (or other auth method) are correctly configured in your environment.
• For large results, adjust execute_sql parameters: timeout_seconds and row_limit to balance latency and data returned.
• The npm_package field is populated when this MCP server is distributed as an npm package; in this case, the package name referenced for startup is databricks-mcp-server.
• The server exposes a predictable set of tools: list_catalogs, list_schemas, list_tables, execute_sql, list_warehouses; you can compose calls to navigate the Databricks metadata and run SQL queries directly.