General-purpose AI assistants can access data, but access alone is not business understanding. To answer questions reliably, an agent needs approved definitions, trusted relationships, authoritative sources, and permission-aware access. Genie One MCP brings that governed context to MCP-compatible assistants and coding agents.
Key highlights
- Define business meaning once: Genie Ontology combines governed semantic assets with context inferred from trusted tables, queries, dashboards, notebooks, and Genie Agents. Genie can resolve business terms, metric definitions, relationships, and source authority instead of asking every client to infer them from raw tables.
- Bring Genie into the tools teams already use: MCP clients such as Claude, ChatGPT, Cursor, and other compatible assistants can send natural-language questions to Genie One. Genie searches permitted enterprise data, generates and runs SQL, and returns grounded answers with links to Databricks sources.
- Keep permissions attached to the user: Unity Catalog permissions are enforced on every request. On-behalf-of user authentication is recommended when answers should reflect each personās access, so two users can ask the same question and receive appropriately scoped results.
- Return more than text: Clients that support MCP Apps can render progress, visualizations, results, and Genie Ontology citations in an interactive view. Clients without MCP Apps support continue to receive text responses.
- Govern the integration centrally: The current Genie One MCP Service, is governed through Unity Gateway. Unity Catalog grants control who can invoke it, service policies can govern tool calls, and invocations are recorded for usage and audit.
- Scope specialized use cases when needed: If a workflow maps to one curated domain, use the Genie Agent MCP server to query a single read-only Genie Agent with its own instructions, trusted SQL, and focused data context.
š Read the full post here
