What Is an MCP Server? How to Use One With Any AI

AI assistants are smart, but on their own they can't see your files, check your tools, or look things up in your apps. MCP fixes that. This guide explains what MCP is, what an MCP server does, and how to add one to four popular AI tools: Claude, ChatGPT, Cursor, and VS Code with GitHub Copilot.

What is MCP?

MCP stands for Model Context Protocol. It's an open standard that lets AI apps connect to outside tools and data in one common way.

Think of it like a USB port for AI. Before USB, every device needed its own cable. With MCP, any AI app that "speaks MCP" can plug into any tool that "speaks MCP," so each new connection doesn't need custom code.

What is an MCP server?

An MCP server is a small program that gives an AI app access to one specific thing, like your files, a database, GitHub, or a search engine.

The AI app asks the server, "What can you do?" The server answers with a list. Then the AI can use those abilities when you ask. For example, a GitHub MCP server might let your AI list open pull requests or create an issue.

Because MCP is an open standard, one server can work with many different AI apps. There's no separate "Claude server" or "ChatGPT server". It's just an MCP server.

How does an MCP server work?

There are three parts:

  • Host: the AI app you use, such as Claude, ChatGPT, Cursor, or VS Code.
  • Client: a connector inside the host. The host makes one client for each server.
  • Server: the program that provides the tools or data.

A server can offer three kinds of things:

  • Tools: actions the AI can take, like "search the web" or "create a file."
  • Resources: information the AI can read, like a file or a database schema.
  • Prompts: ready-made templates for common tasks.

Local vs remote servers

  • Local (stdio): the server runs on your own computer, started by a command. Good for things like your files.
  • Remote (Streamable HTTP): the server runs online at a web address, often ending in /mcp. It usually needs a sign-in (OAuth) or an API key.

This matters because not every AI tool supports both. ChatGPT, for example, only connects to remote servers.

How to add an MCP server to your AI tool

Menus change often, so if a step doesn't match what you see, check the vendor's docs linked at the end.

Claude Desktop

  1. Click the Claude menu in your computer's menu bar (not the settings inside the chat window) and choose Settings…
  2. Open the Developer tab and click Edit Config. This opens claude_desktop_config.json.
  3. Add your server. Here is the official filesystem server:
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/Desktop"]
    }
  }
}
  1. Save, fully quit Claude Desktop, and reopen it.
  2. In the chat box, open the + menu ("Add files, connectors, and more"), hover over Connectors, and choose Manage connectors to see your server.

Claude asks before it uses a tool, so you stay in control.

Claude Code

  1. Open a terminal in your project.
  2. Add a remote server:
claude mcp add --transport http notion https://mcp.notion.com/mcp

Or a local one (everything after -- is the command that starts it):

claude mcp add --transport stdio myserver -- npx -y some-mcp-server
  1. Run claude mcp list to check it's connected.
  2. Inside Claude Code, type /mcp to sign in to servers that need it.

ChatGPT

ChatGPT connects to remote servers only, and it depends on your plan. Full custom MCP support is for Business and Enterprise/Edu workspaces on ChatGPT web (not mobile). On those plans, admins (or members they allow) add servers. Pro users can connect custom servers with read/fetch permissions only.

  1. In ChatGPT on the web, go to Plugins.
  2. Click the + button, then Add custom MCP server.
  3. Enter a name, then your server's public HTTPS URL under Connection.
  4. Set up the sign-in method your server uses.
  5. Read the risk warning, tick I understand and want to continue, and click Create as a plugin.
  6. Start a new chat, type @, and pick your new plugin.

There's no config file for ChatGPT. Everything is done in the menus.

Cursor

  1. In your project, create a file named .cursor/mcp.json. (Use ~/.cursor/mcp.json in your home folder to use the server in every project.)
  2. Add your server. A local one:
{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["-y", "mcp-server"]
    }
  }
}

Or a remote one:

{
  "mcpServers": {
    "remote-server": {
      "url": "https://api.example.com/mcp"
    }
  }
}
  1. Open Customize in the sidebar to check the server is turned on.
  2. Ask for something in chat. Cursor asks for approval before using an MCP tool.

You can also install servers with one click from the Cursor Marketplace in Customize.

VS Code with GitHub Copilot

  1. Open your project folder in VS Code with AI features turned on.
  2. Open the Command Palette (Ctrl+Shift+P, or ⇧⌘P on Mac) and run MCP: Add Server.
  3. Choose HTTP if you have a web address, or Command (stdio) if you have a command, such as npx -y @playwright/mcp@latest.
  4. Give it a short ID, like playwright.
  5. Pick where to save it. For just you, choose Global, then Copilot Global.
  6. Accept any trust or sign-in prompts you understand, then ask Copilot to use a tool from that server.

You can also search @mcp in the Extensions view to browse servers. VS Code still reads the older .vscode/mcp.json file, but now labels it deprecated, so the guided MCP: Add Server flow is the easier route.

How to choose a good MCP server

Before you connect one, check:

  • Is it online? A remote server that's often down will break your work.
  • Is it maintained? Look for recent updates and an active publisher.
  • How does it sign in? OAuth or a limited API key beats pasting a powerful password.
  • What can it access? Give it the least access possible, like read-only database logins or just the folders it needs.
  • Do you trust it? Servers that pull in outside content can carry prompt-injection tricks, so stick to trusted publishers and review actions before approving them.

Find MCP servers on MCP Lantern

MCP Lantern is a directory of MCP servers. It runs regular automated status checks (every few hours), so you can see whether a server was recently up before you connect it. Server owners can list their servers for free, and AI agents can search the directory through MCP Lantern's own MCP endpoint.

If you used Smithery's hosted URLs, note that Smithery's hosted proxy shuts down on October 31, 2026, so it's worth finding each server's direct URL.

Browse MCP servers on MCP Lantern →

FAQ

What does MCP stand for?
Model Context Protocol, an open standard for connecting AI apps to outside tools and data.

Which AI tools can use MCP servers?
Many. Claude Desktop, Claude Code, Cursor, and VS Code with GitHub Copilot all support them. ChatGPT supports remote servers on some plans.

Can ChatGPT use a local MCP server?
Not directly. ChatGPT connects to remote servers over HTTPS. Full support is for Business and Enterprise/Edu plans on the web; Pro users get read/fetch only.

Do I need to code to use an MCP server?
Not usually. Most need one command, a menu option, or a short block of settings. Building your own does take code.

Are MCP servers safe?
They can be. Connect only servers you trust, give them the least access they need, and review actions before approving.

What's the difference between a local and a remote MCP server?
A local server runs on your computer over stdio. A remote server runs online over Streamable HTTP, usually with a sign-in or API key.

Published · More posts