DatabricksMCP

DatabricksMCP is an open-source, safety-first Model Context Protocol server for Databricks. It is being built as a deploy-anywhere gateway for governed access to Databricks workspaces from AI assistants and agents.

Important

This project is under active development and is not yet ready for production. The current Phase 1 surface is deliberately read-only.

DatabricksMCP is an independent community project. It is not affiliated with, endorsed by, or sponsored by Databricks, Inc. Databricks and the Databricks logo are trademarks of Databricks, Inc.

Phase 1 capabilities

• MCP over local stdio and remote Streamable HTTP
• Databricks unified authentication through the official Python SDK
• Health and authenticated-identity diagnostics
• Read-only Unity Catalog and SQL warehouse discovery
• A deny-by-default policy foundation for future mutation tools
• Typed configuration, structured errors, tests, CI, and a non-root container

See ROADMAP.md for the planned production-grade capability set.

Quick start

Requirements:

• Python 3.11 or newer
• A Databricks workspace
• Databricks unified authentication configured through environment variables or a profile in ~/.databrickscfg

Install locally with uv:

uv sync --all-groups
uv run databricks-mcp

Configure a local MCP client:

{
  "mcpServers": {
    "databricks": {
      "command": "uvx",
      "args": ["--from", "databricks-mcp-server", "databricks-mcp"]
    }
  }
}

The package is not published yet, so during development point the client at a local checkout:

{
  "mcpServers": {
    "databricks": {
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "/absolute/path/to/DatabricksMCP",
        "databricks-mcp"
      ]
    }
  }
}

Authentication

DatabricksMCP delegates credential discovery and refresh to Databricks unified authentication. For local development, authenticate with the Databricks CLI or select a named profile:

export DATABRICKS_CONFIG_PROFILE=DEFAULT
uv run databricks-mcp

For unattended environments, use OAuth M2M or workload identity federation. Personal access tokens are supported by the SDK but are not recommended for production deployments.

Never place credentials in MCP tool arguments or commit them to configuration.

Streamable HTTP

export DATABRICKS_MCP_TRANSPORT=streamable-http
export DATABRICKS_MCP_HOST=127.0.0.1
export DATABRICKS_MCP_PORT=8000
uv run databricks-mcp

The HTTP endpoint is /mcp. Phase 1 HTTP mode is suitable for local evaluation; public deployment must be placed behind standards-compliant authentication and TLS until native remote authorization lands in Phase 3.

Configuration

Variable	Default	Purpose
DATABRICKS_MCP_TRANSPORT	stdio	stdio or streamable-http
DATABRICKS_MCP_HOST	127.0.0.1	HTTP bind address
DATABRICKS_MCP_PORT	8000	HTTP bind port
DATABRICKS_MCP_LOG_LEVEL	INFO	Python log level
DATABRICKS_MCP_ACCESS_MODE	read-only	Policy mode; only read-only exists in Phase 1

Standard DATABRICKS_* authentication variables are consumed by the official Databricks SDK.

Development

uv sync --all-groups
uv run ruff check .
uv run ruff format --check .
uv run mypy src
uv run pytest

Architecture and security decisions live in docs/. Contributions are welcome; read CONTRIBUTING.md and SECURITY.md before opening an issue or pull request.

License

Licensed under the Apache License 2.0.