What is MCP (Model Context Protocol)?

The Model Context Protocol (MCP) is an open standard that gives AI agents a single, consistent way to reach tools and data. Instead of writing a bespoke integration for every service an agent needs, you point the agent at an MCP server and it discovers what that server can do. One protocol, many servers.

Why MCP exists

Before MCP, every agent framework invented its own "tools" format. Connecting an agent to GitHub, a database, or a search API meant custom glue code that only worked inside that one framework. That glue multiplied: N agents times M tools is a lot of one-off integrations to build and maintain.

MCP collapses that to N + M. A tool author writes one MCP server; any MCP-speaking agent can use it. An agent author adds MCP support once; every MCP server becomes available. Our MCP server directory catalogues servers you can plug in this way.

The three things a server exposes

An MCP server offers up to three kinds of capability:

• Tools — actions the agent can invoke (create an issue, run a query, send a message).
• Resources — data the agent can read (files, records, documents).
• Prompts — reusable prompt templates the server suggests for common tasks.

The agent connects, asks the server what it offers, and calls those capabilities as the task requires. Transport is typically stdio (a local subprocess) or HTTP/SSE (a remote server).

What MCP does not do

MCP is a connection standard, not a runtime. It does not decide when an agent should call a tool, host your agent, or manage memory — that is the job of the agent and the platform running it. MCP just standardises the wire between the two. See our glossary for how these pieces fit together.

Why it matters when choosing a platform

Because MCP is where the ecosystem is consolidating, how well a platform supports it is a real differentiator. Some platforms are MCP-native — they treat MCP as a first-class way to add capabilities. Others bolt it on partially or not at all. We record an MCP support fact (native / partial / none) on every platform dossier, and rank the best MCP-native platforms by that support.

The distinction is significant enough that we wrote a dedicated guide on it: MCP-native vs bolt-on.

Where to go next

• Connect one: How to connect an MCP server to your AI agent walks through the practical steps.
• Browse servers: the MCP directory lists servers by category.
• Compare platforms: AI agent platforms, compared records MCP support alongside pricing, hosting, and memory for each option.

MCP is young but moving fast. If you are choosing tooling today, favouring MCP-native pieces keeps your options open as the standard settles.