What is RAG vs MCP?

RAG (Retrieval-Augmented Generation) and MCP (Model Context Protocol) are often compared, but they operate at different levels and are complementary rather than competing. RAG is a pattern: take a user's query, search a knowledge base, typically a vector database holding embedded documents, and paste the top matches into the model's context so its answer is grounded in your data. MCP is a connectivity standard: it defines how a host and model talk to external systems through tools, resources, and prompts, regardless of what those systems do. The clearest way to see the relationship is that RAG can be delivered over MCP. You can wrap a vector store and its semantic search behind an MCP server that exposes a search tool; now any MCP-aware agent can perform RAG against that store. MCP also goes beyond retrieval, it lets the model take actions (open a PR, send a message), maintain memory, and chain tools, which plain RAG does not address. In short: reach for RAG when the problem is grounding answers in a document corpus, and think of MCP as the universal interface through which that retriever, and many other capabilities, are exposed to agents.