Home / Articles

What Is an MCP Server? Definition, Examples & How to Use One

An MCP server gives AI apps like ChatGPT, Claude and Cursor tools they can call: read files, query data, run commands. Plain-English definition, how it works, examples, local vs remote, and how to use one.

When you chat with ChatGPT or Claude, the AI can only work with what is inside the conversation. It cannot open a file on your laptop, look at your database or check your server. The Model Context Protocol (MCP) changes that, and an MCP server is the piece that makes it happen.

MCP server: the short definition

An MCP server is a program that offers tools (and optionally data and prompts) to AI apps through the Model Context Protocol, an open standard. A tool can be anything with clear inputs and outputs: read a file, search documents, query a database, create a calendar event or run a command. The AI app that connects to the server and uses those tools is the MCP client.

An everyday comparison

Think of USB. Before it, every device needed its own cable. With one standard plug, any device works with any computer. MCP does the same for AI: a tool built once as an MCP server can be used by ChatGPT, Claude, Cursor and other apps that support MCP, instead of a separate plugin for each.

How an MCP server works

  1. Connect. The AI app (client) connects to the server and asks which tools it offers. Each tool comes with a name, a description and the inputs it expects.
  2. Decide. When you ask for something, the AI model reads the tool descriptions and decides whether a tool would help, for example read_file with a path.
  3. Call. The app sends the tool call to the server. Messages use JSON-RPC.
  4. Answer. The server does the work and returns the result, which the AI uses to answer you or to decide the next step.

Many apps ask you to confirm before a tool changes something, so you stay in control of what actually happens.

Local vs remote MCP servers

  • Local MCP servers run on your own computer and talk to an app on that same computer, usually over standard input/output. Good for desktop apps and coding tools.
  • Remote MCP servers run on the internet and are reached over HTTPS (the "streamable HTTP" transport), usually with OAuth sign-in. Web and mobile apps such as ChatGPT in the browser or Claude on your phone can only use remote servers.

Shellway is a remote MCP server with a twist: it relays each request to a small app on your own computer, so web and mobile AI apps can work with your real files and terminal without you opening ports.

Examples of MCP servers

  • Files and terminal on your computer: read, edit and search files, run commands (for example Shellway, or local filesystem servers).
  • Code hosting: read issues and pull requests, review code.
  • Databases: look up tables and run read-only queries.
  • Productivity apps: calendars, documents, task trackers, email.
  • Servers and DevOps: check logs, restart a service, deploy (see how developers use MCP to deploy and monitor servers).

Which apps support MCP servers?

Support keeps growing. Today you can use MCP servers in ChatGPT (as connectors and plugins), Claude (claude.ai, desktop and mobile, and Claude Code), Cursor, VS Code, Gemini CLI and Codex, among others. Remote servers work in the web and mobile apps; local servers work in desktop apps and coding tools.

How to use an MCP server

  1. Pick a server for the job, and check who runs it and what its tools can do.
  2. Add it to your AI app: from the app's plugin or connector directory, or as a custom connector with the server URL (for example https://mcp.getshellway.com/mcp).
  3. Sign in when asked (OAuth), and turn the connector on in a chat.
  4. Ask in plain words. The AI decides when to call the tools.

Step-by-step guides: connect ChatGPT to your computer, Claude, Cursor.

Is it safe to use an MCP server?

An MCP server can only do what its tools allow, and you decide which servers to connect. Before adding one, check:

  • Who runs it, and whether it has a clear privacy policy.
  • Which tools can change things (write, delete, run) and whether there is a read-only option.
  • Whether you can see a log of what was done, and disconnect at any time.

With Shellway you also choose exactly which computers are connected, you can run the app in read-only mode, and every action is logged by time, tool and result.

Common questions

Is MCP only for developers?

No. Building a server takes some code, but using one is like adding an app: connect it, sign in and ask in plain words.

What is the difference between an MCP server and an API?

An API is built for programs that already know how to call it. An MCP server describes its tools in a standard way so an AI model can discover them and decide when to use them, and one server works across many AI apps.

Do I need to keep my computer on?

For a server that works on your own computer, like Shellway, yes: the computer must be on and online. Cloud-based MCP servers run independently.

Try it

See the setup guides for ChatGPT, Claude and Cursor, or browse the prompt library for ideas.