AI icon

MCP Explained: The Protocol Making AI Agents Actually Useful

3 min readByArnau Ruiz

MCP is yet another three-letter acronym in a field that already has plenty, but this one is actually worth learning. Model Context Protocol is the thing that lets an AI assistant stop being a chatbot that just talks about your server, and start being one that can actually read your files, run commands, or update a WordPress post. If that sentence made you nervous, good, it should, but it is also genuinely useful once you understand what it is and is not doing.

What MCP actually is

MCP is an open protocol that defines a standard way for an AI model to talk to external tools and data sources, things like your file system, a database, a Git repo, or in this blog’s case, the WordPress install running under it. Instead of every AI product inventing its own one-off integration for every tool, MCP gives both sides a common language: the tool exposes a small server describing what it can do, and the AI client connects to it and calls those capabilities.

Think of it less like magic and more like a very well-documented API that an AI model happens to be the one calling.

The two halves: server and client

  • An MCP server exposes specific capabilities, reading a file, querying a database, running a WordPress function, and describes them so an AI model can understand what each one does and what input it needs
  • An MCP client, usually the AI assistant’s own app or interface, connects to one or more of these servers and decides, based on what you ask it to do, which capability to call and with what arguments
  • The AI model itself never gets raw, unrestricted access to your system, it only gets whatever the server explicitly decided to expose

Why this matters more than the acronym suggests

Before something like MCP existed, connecting an AI assistant to a real tool meant custom, one-off integration work for every single combination of AI product and tool. MCP turns that into a many-to-many problem instead of a one-to-one one: build one MCP server for your tool, and any MCP-compatible AI client can use it, no bespoke integration required on either side.

This is exactly the mechanism behind an AI assistant managing a WordPress site: an MCP server exposes specific WordPress capabilities, and the AI client calls them one at a time, the same way it would call any other tool.

A minimal example

An MCP server’s capability list is just structured data describing what it offers, something along these lines:

{
  "name": "get_weather",
  "description": "Get current weather for a city",
  "parameters": {
    "city": "string"
  }
}

The AI model reads that description, decides when it’s relevant to a request, and calls it with the right parameters. Multiply that by dozens of tools and you get an AI assistant that can genuinely act, not just describe.

Should you actually care about this

If you’re a sysadmin or homelabber, yes, quietly. MCP is becoming the way self-hosted tools, dashboards, and automation platforms let AI assistants interact with them, which means an increasing number of the tools already on this blog (Portainer, Uptime Kuma, your WordPress install) either have or will have an MCP server available. Worth knowing the term exists before it shows up in a plugin’s changelog with no explanation.

Related reading