You'll see "MCP server" mentioned in almost every AI coding tool now. It sounds technical, but the idea is simple. This guide explains it in plain English, with no coding required to follow along.
The short version
MCP stands for Model Context Protocol. The official site describes it as an open-source standard for connecting AI applications to external systems, and compares it to a USB-C port for AI: one common plug that fits many things.
An MCP server is a small program that offers something to your AI app, such as access to your files, a database, GitHub or a web browser. Without one, an AI assistant can only work with what you type or paste in. With one, it can look things up or take actions in other tools.
The three parts
The official docs use three terms:
| Term | What it means | Example |
|---|---|---|
| Host | The AI app you use | Claude Code, Claude Desktop |
| Client | The part inside the host that keeps a connection to one server | You don't see it. The host makes one per server |
| Server | The program that provides the extra abilities | A GitHub server, a file server |
So in day-to-day use you only deal with two things: the app you chat with, and the servers you plug into it.
What can a server offer?
According to the MCP docs, servers can provide:
- Tools: actions the AI can ask to run, such as "create an issue".
- Resources: data the AI can read, such as the contents of a file.
- Prompts: ready-made templates for common jobs.
Most beginners will mainly notice the tools.
How do clients connect?
There are two common styles:
- Local servers run on your own computer. The app starts the program itself and talks to it directly. The docs call this the stdio transport.
- Remote servers live on the internet and you connect by web address. The docs call this Streamable HTTP. Remote servers often ask you to sign in.
How you add a server depends on the app:
- Claude Code uses a command. The docs show this shape for a local server:
claude mcp add [options] <name> -- <command> [args...]
and this one for a remote server:
claude mcp add --transport http <name> <url>
See Connect Claude Code to MCP servers for a full walkthrough.
- Cursor is set up through its settings. See Add an MCP server to Cursor.
- Claude Desktop has a friendlier route for non-coders. See Claude Desktop connectors for non-coders.
Once connected, you just chat as normal. The AI decides when a server's tools would help, and the app typically asks your permission before it uses them.
Where to find good servers
- Our directory. Browse the PromptKitchen MCP directory for servers with install notes for the clients we cover, for example Filesystem and GitHub.
- The official reference servers. The modelcontextprotocol/servers repository lists a few maintained examples. Its README calls them reference implementations and says they are not production-ready solutions.
- The official MCP Registry. The same README points to the MCP Registry, where published servers can be browsed.
Staying safe
An MCP server can read data and take actions on your behalf, so treat each one like software you're installing:
- Only add servers you trust. Claude Code's docs say to verify you trust each server before connecting it, and warn that servers which fetch outside content can expose you to prompt injection (hidden instructions in web pages or documents that try to steer the AI).
- Give the least access needed. If a server asks for a folder, a repository or an account, choose the narrowest option.
- Never paste secrets into shared config files. Don't commit API keys or tokens to Git.
- Read permission prompts. Approve actions that change or delete things only when you understand them.
- Remove what you don't use. Fewer servers means fewer things to worry about.
Is it right for you?
If you only chat with an AI about general questions, you probably don't need MCP yet. It earns its place when you want the AI to work with your real tools: your code on GitHub, files on your computer, a browser to test your site.
A good first step is a single, simple server. Add it, try a few requests, then decide whether you want more.