The Databricks MCP Server provides an agent-facing interface to Databricks Unity Catalog metadata and data lineage. It is designed to empower LLMs to autonomously understand data assets (catalogs, schemas, tables, columns), inspect data processing code (notebooks and jobs), and craft SQL queries against a Databricks SQL warehouse. Core capabilities include tools to list catalogs, describe catalogs and schemas, describe tables with optional column details and lineage, and to execute SQL queries via the Databricks SDK. In agent mode, the server supports iterative exploration—discovering sources, inspecting table structures, analyzing upstream/downstream lineage, and inspecting notebooks for business logic and data quality checks—culminating in query construction. It also provides a Markdown-friendly output format for LLM consumption. The server supports metadata-as-code workflows with Terraform examples to define Unity Catalog assets and a Terraform-import workflow. For long-running queries, a 50-second wait timeout is documented. Standalone operation via main.py and integration with Cursor are supported. Dependencies include databricks-sdk, python-dotenv, mcp[cli], asyncio, and httpx.