S&P Global Energy, an energy and commodity data provider, says subject-matter experts can now publish conversational access to governed data for AI agents without engineers building a new interface for each domain. The company built the system with Databricks Genie Agents and the Model Context Protocol (MCP).
S&P Global Energy’s structured data spans chemicals, crude oil, refined products, gas and power, and liquefied natural gas. Each dataset group has its own Genie Agent, curated by subject-matter experts who add table descriptions, example queries and business definitions rather than writing code.
Those agents are then exposed as MCP servers. Databricks’ Unity Catalog applies the existing permissions around the underlying data, while a shared proxy layer groups the domain-specific agents so a calling agent can send a query to one or several data domains.
The change shortens a process that previously ran through requirements, API design, engineering and testing. Priyanka John, vice president at S&P Global Energy, said work that previously required a full development cycle can now be completed in days.
Engineering moves up a layer
Engineering has not disappeared from the process. Its role has shifted from building a separate conversational interface for every data domain to maintaining the common infrastructure that connects and combines those domain-specific agents.
The same endpoints can also reach outside S&P Global Energy. The companies said customers can connect their own MCP-compatible agents and assistants directly to governed S&P Global Energy data, with Unity Catalog permissions applied to those requests.
Data reaches the customer’s agent
This allows S&P Global Energy to reuse the same governed access layer across customer AI environments rather than build a separate connection for each application. Databricks lists its managed MCP servers, including Genie Agent servers, as Public Preview.
The architecture also changes how additional conversational data products are created. Instead of beginning each new domain with another engineering cycle, S&P Global Energy can reuse the same access layer while subject-matter experts prepare individual datasets for agent queries.
S&P Global Energy said it measures accuracy through Genie Agent Benchmarks and reruns those evaluations after changes to the data or agent instructions. The joint announcement did not disclose benchmark scores, measured deployment-time data, or figures on external-customer adoption.