# WHAT IS MCP (MODEL CONTEXT PROTOCOL)?

> MCP turns the N-by-M integration problem into N plus M. Write a server once and any client that speaks the protocol can use your tools — which is why it spread so quickly.

- Canonical: https://www.alpacamango.com/docs/what-is-mcp-model-context-protocol
- Author: Alpaca Mango
- Published: 2026-08-10 · Updated: 2026-08-09
- Topics: AI Agents, Ai, Architecture

**MCP — the Model Context Protocol — is an open standard for connecting AI applications to external tools and data sources.** A server exposes tools, resources and prompts; any client that speaks the protocol can use them. Its purpose is to stop everybody writing the same integration once per framework.

## The problem it solves

Without a protocol, connecting *M* AI applications to *N* data sources is an *M × N* problem: every application needs bespoke code for every source, and every source's authors get asked for one integration per popular client.

With one, it is *M + N*. Write a server once, and every compliant client gets it. That is an unglamorous, structural improvement, and it is the entire reason adoption moved as fast as it did — the appeal is not the technology, it is the arithmetic.

## The three things a server exposes

- **Tools** — functions a model may call, with described arguments. The part that looks like ordinary tool calling.
- **Resources** — data a client can read into context: files, records, query results.
- **Prompts** — reusable templates a server offers, so the people who know the domain can ship the wording rather than describing it in a README.

## Why it changes framework choice

It lowers the stakes. Historically, a large part of a framework's value was its integration catalogue — the reason to accept [LangChain](/docs/langchain-vs-langgraph)'s abstractions was that it already talked to everything. When integrations live behind a shared protocol, that moat shrinks, and you can pick a framework on the merits of its control flow instead.

Which makes the smaller, more opinionated options — [Pydantic AI](/docs/pydantic-ai-vs-pydantic), the [OpenAI Agents SDK](/docs/openai-agents-sdk-explained) — more viable than they would otherwise be, and strengthens the case for [no framework at all](/docs/agent-framework-or-no-framework).

## What to be careful about

An MCP server is code you are granting access to your data and your tools, and "it is just a config entry" is exactly the framing that gets one installed without review. Treat adding a third-party server as adding a dependency with credentials — because that is what it is. Prefer servers you can read, scope their permissions narrowly, and be as suspicious of tool descriptions from an untrusted server as you would be of any other untrusted input reaching a model.