RAG-Ready Patterns for Data Platforms

Chapter 8. Orchestrating RAG with the Model Context Protocol

Data platforms were already complex before LLMs hit the mainstream. A typical enterprise data platform spans ingestion services, pipelines, semantic models, quality monitors, cost and governance tracking systems, reporting dashboards, and more, each with its own APIs, tools, and workflows. Up to this point, we focused on getting the data platform “RAG-ready” by curating high-quality knowledge, exposing reliable metrics, and tightening governance. In this chapter, we focus on the next step: how to enable AI systems to use those capabilities on behalf of your users.

The Model Context Protocol (MCP) is the interface that makes this possible. MCP gives models a standard, declarative way to discover and invoke tools, services, and data products on behalf of users, without baking platform-specific assumptions into prompts or bespoke code. The orchestration (i.e., deciding which tools to call, in what order, and what to do with the results) belongs to the agent. MCP’s contribution is to make the available capabilities describable, invocable, and governable in a uniform way.

Instead of hiding logic inside opaque agent frameworks, you expose your platform as a set of MCP tools and let agents compose them into explicit, standardized retrieval and action chains in response to user intent. MCP provides the uniform invocation surface; the transparency and auditability that make those chains trustworthy come from the instrumentation, ...