How does MCP differ from RAG, and when would you choose one over the other?

MCQ·Medium·4.0 · 0·~1 min·Relevant atAnthropic

Attempt it

Quick reference

TL;DR

RAG is a retrieval pattern that injects documents before generation; MCP is a connection protocol for tools and data. They are different layers, and MCP can expose RAG as one tool.

Memory aid

Easy to grasp

Picture a chef in a kitchen. RAG is like prepping ingredients before service: you gather the right items and lay them on the counter so the chef has them on hand while cooking. MCP is the standardized plumbing and power sockets in the kitchen: it lets the chef plug in any appliance, a blender, an oven, a scale, from any brand without rewiring. One is a way of getting facts ready before you start. The other is a wiring standard that lets the chef reach live tools and data while working. You can even build a 'fetch the right ingredients' appliance and plug it in through that standard plumbing. So they are not rivals; one can sit inside the other.

Key concepts

LLM tool use Retrieval-augmented generation AI agents Retrieval pattern

Detailed answer & concept explanation

~7 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.

Quick scan · 2–4 min

Everything important, quickly.

4 min: RAG as a retrieval pattern + MCP as a connection protocol + the containment relationship + when to pick each + how they compose in a real agent.

Side-by-side

Concern	RAG	MCP
What it is	Retrieval pattern	Connection protocol
When it runs	Once, before generation	At runtime, mid-conversation
Primary job	Ground the answer in a corpus	Wire the app to tools, data, prompts
Transport	None; just prompt assembly	JSON-RPC over stdio or streamable HTTP
Containment	Cannot wrap MCP	Can expose RAG as one tool

Real-world examples

Real products, models, and research that use this idea.

A Claude Code session can register an MCP server whose search tool runs a vector-index RAG lookup over a private repo.
Anthropic's official MCP servers expose Postgres and filesystem access, which an app can layer RAG on top of.
Cursor connects to MCP servers for live tool access in 2026 while still running classic RAG over indexed project docs.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QHow would you decide between up-front RAG injection and a runtime MCP retrieval tool for the same corpus?

QIf an MCP server exposes a retrieval tool, where do the classic RAG design choices still live?

QWhat failure modes appear when you put RAG behind an MCP tool that did not exist with plain RAG?

Don't say this

Red flags and common mistakes that signal junior thinking. Click to expand.

Most common mistake

Treating MCP as a newer replacement for RAG. They solve different problems: RAG is a retrieval pattern, MCP is a connection protocol that can carry a retrieval tool.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

References

Primary sources. Skim if you want the original framing.