MCP: إيه هو وليه كل أدوات الذكاء الاصطناعي بتدعمهMCP: what it is and why every AI tool supports it

MCP stands for Model Context Protocol, an open standard that defines one way for AI applications to connect to tools and data outside the model: your files, your database, GitHub, the browser, anything. Instead of building a custom connector per tool per app, you write it once as an “MCP server” and it works with any app that supports the protocol.

The problem it solves

A language model on its own is sealed. It knows what it was trained on and what you typed into the conversation, and that is all. To read a file, query a database or open a ticket, something has to hand it a “tool”.

Before MCP every app defined tools its own way. With 5 apps and 10 services you needed 50 connectors. Anthropic announced MCP in November 2024 as an open standard so the sum becomes 5 + 10: each service has one server, and every app speaks the same language.

The usual comparison is USB-C: one port, and anything plugs in.

How it works

There are three parties:

• The host: the app you use that contains the model, such as a code editor or a chat app.
• The client: a part of the app that opens a connection to each server.
• The server: a small program that exposes the abilities of one service.

A server can expose three kinds of things:

• Tools: actions the model can run, like “search files” or “run a query”.
• Resources: data to read, like a file’s content or a table row.
• Prompts: ready-made instruction templates the user can pick.

The two sides talk in JSON-RPC messages. A server can run locally on your machine and talk over stdio, or live on the internet and talk over HTTP.

How it differs from a normal API

An API is made for a programmer who reads the docs and writes code that calls it. MCP is made for a model that decides at run time what to call.

That is why every MCP tool carries a plain-language description and the shape of its inputs. The model reads that description, understands what the tool does, and decides whether to use it. Most of the time an MCP server is a thin layer over an API that already exists.

Example: adding a server

Most apps read their servers from a JSON settings file. This example adds the official filesystem server and allows it a single folder:

{
  "mcpServers": {
    "files": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/me/projects/site"
      ]
    }
  }
}

The file’s name and location differ between apps, so check the docs of the one you use. After a restart, the model sees tools such as read file, write file and search the folder, and uses them when your request needs them.

Security: the part people forget

An MCP server is a program running with your permissions that gives a model the ability to act. So:

• Do not install a server from a source you do not trust. It is the same as any program you download and run.
• Least privilege. One folder, not the whole machine. A read-only key if you do not need writes.
• Watch for prompt injection. A web page, a file or an email can contain text written for the model, meant to make it do something you never asked for. What the model reads is data, not orders.
• Review dangerous actions. Deleting, sending, paying: make them wait for your approval.

Every tool you give the model is a permission you give to any text it will read.

Where to start

1. Pick one app you really use that supports MCP.
2. Add one server for something you do daily, like files or GitHub.
3. Use it for a week before adding another. Many servers mean many tools, and the model gets confused.
4. When you need something that does not exist, write your own server. There is an official SDK for several languages.

For the bigger picture of this way of working, read Vibe coding: what it is and when to trust it. And if you want to connect your tools and automate your work step by step, that is the subject of the AI automation course.