MCP on raw data is just Excel hell again

AI agents connected directly to raw systems through Model Context Protocols (MCPs) can quickly answer business questions but may produce inconsistent results across departments because they lack centrally governed definitions and coded business rules. The piece compares this problem to the pre-data-warehouse era, when analysts and teams created separate spreadsheet-based calculations, while arguing that agents replace the analyst’s semantic interpretation role rather than the warehouse’s function as a precise, tested source of truth. It recommends separating a meaning layer, such as an ontology or business dictionary defining metrics like customers and revenue, from a precision layer containing tested code that calculates those metrics consistently across data sources. Agents should access governed tools mapped to these layers rather than raw operational data, particularly because raw MCP connections may also permit harmful write actions. The article cites semantic and governance capabilities from Databricks and Snowflake as examples, and presents dltHub as a tool intended to build pipelines, models, ontologies, and metric views within existing platform governance systems.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.