What is MCP ?

MCP is the method by which AI Tools can be hosted on the internet and used by your AI Agents.

AI Tools can either be used locally or hosted on the internet.

When hosted on an API, we need some way to tell our AI which tools are available, and how to use them. If we did this ourselves, we would either have to explain every tool, how it's used, and when to use them, or we could copy it from somewhere and paste it into our AI prompt. This could really quickly become annoying.

Who better to describe how a tool works, how to use it and when, than the person who created it ?

In the world of AI, MCP is a way to do exactly that. It allows the AI tool itself to detail its purpose, all the tools it contains, and how to use them. That way, whenever the tool builder adds more tools or changes anything, our AI will immediately know about it.

MCP is similar to SEO. SEO is a way for websites to tell search engines (like Google) what is available, so when you search it can recommend those websites if they are relevant. MCP does the same for AI tools. If you ask your AI to schedule a calendar event, it will check all the tools it has available, find the best one, and use it. MCP is a way for the tool to tell your AI what it can do and when it should be used.

MCP exists in two parts. Servers and Clients. An MCP Server is where the tools are hosted. A Client is where the tool is used. Your AI Chat is the MCP Client. The MCP you connected to is known as the Server.

MCP is just a way to host AI Tools over an internet connection, rather than running them locally.

MCP is the method by which AI Tools can be hosted on the internet and used by your AI Agents.