How can ChatGPT, Claude, and other AI tools use the same governed business definitions?

Summary

• External AI tools reach your governed definitions through an open standard, the Model Context Protocol (MCP), instead of each tool keeping its own private copy of what your metrics mean.
• On Databricks you model definitions once in Unity Catalog and Genie Ontology, then expose them through managed MCP servers so any compliant client can use them.
• An MCP-compatible assistant such as ChatGPT or Claude sends a natural-language question to a Databricks Genie MCP server, which answers from your certified metrics and permissions and returns links back to the sources.
• Requests run under the caller's Unity Catalog permissions, so a tool sees only what its user is allowed to see.
• Genie Ontology is in Public Preview; treat its availability and behavior as subject to change.

How can ChatGPT, Claude, and other AI tools use the same governed business definitions?

They connect to your governed definitions through an open standard, the Model Context Protocol (MCP), instead of each tool keeping its own private copy of what your metrics mean. On Databricks, you model business definitions once in Unity Catalog and Genie Ontology, then expose them through managed MCP servers. An MCP-compatible assistant such as ChatGPT or Claude sends a natural-language question to a Databricks Genie MCP server, which answers using the same certified metrics, glossary terms, and permissions that power Databricks' own surfaces, then returns links back to the sources. Define the metric once, and every client inherits it.

What is MCP, and why does it matter here?

The Model Context Protocol is an open standard for connecting AI applications to tools and data. It matters because it turns governed business context into something any compliant client can call, rather than a capability locked to one product. Databricks publishes managed MCP servers that expose governed capabilities, including a Genie server for natural-language questions over your data. The server URL follows the pattern https://<workspace-hostname>/api/2.0/mcp/genie, and access uses OAuth, so each connection is authenticated and scoped rather than open-ended.

Key capabilities to evaluate

For several AI tools to share one set of definitions safely, look for these properties.

Capability	Why it matters
Single source of definitions	Metrics and terms modeled once, so every tool reports the same numbers.
Open-standard access (MCP)	Any compliant client can reach the definitions without a bespoke integration or lock-in.
Permission inheritance	Each request runs under the caller's identity, so a tool sees only what its user may see.
Central governance and audit	One place to set access, rate limits, and logging across every AI surface and tool.
Grounded, cited answers	Responses trace back to authoritative sources instead of a model's guess.
Conflict resolution	An authoritative definition wins when teams model the same metric differently.

How Databricks approaches shared, governed definitions

The governed source of truth is Unity Catalog semantics plus Genie Ontology. Unity Catalog holds the deliberate definitions: metric views, business glossary terms, and domains you certify. Genie Ontology, in Public Preview, adds context it learns automatically from your dashboards, queries, metric views, and notebooks. It ranks competing definitions by authority with an algorithm called OntoRank, prefers certified definitions when they conflict, and gates every snippet behind Unity Catalog permissions.
Because that context is modeled once, it is reused everywhere. The Databricks Genie experience, AI/BI dashboards, and Genie Agents all draw on it, and so do outside assistants that connect over MCP. When a client such as ChatGPT or Claude calls the Genie MCP server, Genie interprets the question against your ontology, writes SQL, and returns an answer grounded in the definitions your business has certified, with deep links to the underlying Databricks sources. The tools act as clients of the same governed context; each one does not hold a separate copy that can drift.
Governance stays centralized through Unity Gateway, the control plane for AI traffic. It applies access control, rate limits, budgets, auditing, and MCP governance across surfaces, so requests from external tools inherit the same policies as internal ones. One governed definition set plus one control plane is how many tools can share business meaning without adding risk. For a deeper walkthrough, see Operationalizing Genie Ontology.

FAQs

Do external AI tools need a separate integration for each metric?

No. Definitions are modeled once in Unity Catalog and Genie Ontology and exposed over MCP, so any compliant client reaches the same set without per-metric integration work.

Does connecting an outside tool bypass our permissions?

No. Requests run under the caller's identity and Unity Catalog permissions, and Unity Gateway applies access control, rate limits, and audit logging to MCP traffic.

How do the tools stay consistent with our dashboards?

They read the same certified metric views, glossary terms, and ontology that the Databricks Genie experience and AI/BI dashboards use, so the numbers match by construction.

Is Genie Ontology generally available?

Genie Ontology is in Public Preview. Confirm current availability and behavior in the Databricks documentation before relying on it in production.