MCP Databricks Server

What JordiNei/mcp-databricks-server MCP server does

The JordiNei/mcp-databricks-server MCP server exposes selected Databricks operations as MCP tools. It connects to a Databricks workspace and supports SQL execution through a configured SQL warehouse, along with read-oriented job management queries.

The server is intended for LLM clients that support the Model Context Protocol. Typical requests include discovering tables with SQL, counting rows in a table, listing jobs in a workspace, checking a job by numeric ID, or retrieving more complete job configuration and status information.

How it works

The process runs locally from the cloned repository. On startup, it reads connection settings from a .env file in the project directory, then uses those settings to communicate with Databricks. SQL statements are sent to the specified SQL warehouse, while job operations use the Databricks workspace connection.

The repository documents Python 3.7 or newer and recommends creating a virtual environment before installing the dependencies listed in requirements.txt. A separate test_connection.py script is available for checking the Databricks connection before starting the MCP process.

Setup and configuration

Create these environment variables:

• DATABRICKS_HOST: Databricks workspace host without the https:// prefix.
• DATABRICKS_TOKEN: Personal access token for the workspace.
• DATABRICKS_HTTP_PATH: HTTP path for the target SQL warehouse.

The Databricks account must provide a workspace, an accessible SQL warehouse, and permissions to run queries and access jobs. The documented startup command is python main.py. The README also shows an optional MCP Inspector command for testing, but the Inspector is a separate debugging utility rather than part of the server itself.

Tools and capabilities

The server defines four MCP tools:

• run_sql_query(sql: str) executes a SQL statement on the configured Databricks SQL warehouse.
• list_jobs() returns the jobs available in the Databricks workspace.
• get_job_status(job_id: int) retrieves the status of one job using its numeric ID.
• get_job_details(job_id: int) retrieves detailed information for one job.

These tools cover warehouse queries and basic workspace job inspection. The provided material does not describe tools for creating, editing, deleting, or running jobs.

Limitations and notes

The JordiNei/mcp-databricks-server MCP server requires valid Databricks credentials and appropriate workspace permissions. The SQL warehouse must be running and reachable for query execution. A personal access token grants direct access to the workspace, so it should be limited to the required permissions, kept out of version control, and protected through secure environment handling.

The documentation does not identify a published package, hosted endpoint, specific MCP client, or software license. Installation therefore assumes a local repository checkout and Python environment rather than a package-manager deployment.