Databricks MCP

Databricks is a lakehouse platform where tables, functions, indexes, and models are governed through Unity Catalog. Its MCP surface is not one server but a small family of them, each wrapping a different part of the platform, so an agent can ask questions of company data without a bespoke integration for every asset type. The managed servers cover what an agent usually needs. Vector and AI search indexes handle retrieval of relevant documents. Unity Catalog functions are callable as ready made SQL tools, which lets a data team publish a vetted query as an agent capability instead of trusting generated SQL. Genie spaces answer natural language questions over a curated set of tables. A SQL interface runs generated queries against a warehouse asynchronously, submitting the statement and polling until results are ready rather than holding a connection open. Databricks Labs and community servers extend the surface to workspace objects such as jobs, clusters, and notebooks. Governance is the reason to prefer this over a direct JDBC connection. Unity Catalog evaluates the calling identity on every request, so row level filters, column masks, and lineage tracking apply to the agent exactly as they would to a person running the same query. That makes it a realistic way to give an assistant access to warehouse data without carving out a separate, less controlled path around your existing permission model.