What Are MCP Servers? A Plain-English Guide With Examples

Short answer: What are MCP servers? They are small programs that give an AI app access to one outside system, such as your files, GitHub or a database, through a shared standard called the Model Context Protocol. The AI app connects as a client, asks the server what it can do and calls those tools when your request needs them. One server can work with many apps, so nobody has to build the same connection twice.

If you want to see real ones before reading on, our MCP servers directory lists 100 servers in 10 categories, with install steps, sign-in method and price for each. This guide explains the idea behind them in plain words, shows how a request travels from your chat to a server and back, and covers what to check before you install one.

MCP servers at a glance

What is an MCP server?

An MCP server is a program that lets an AI app use one outside system through the Model Context Protocol. The protocol itself is an open-source standard for connecting AI applications to data sources such as local files and databases, tools such as search engines and calculators, and workflows such as specialized prompts.

The official introduction uses a good picture: MCP is like a USB-C port for AI applications. Before USB-C, every device needed its own cable. Before MCP, every AI app needed its own custom code for every service. Anthropic made the same point when it announced the protocol in November 2024: each new data source needed its own custom implementation, which made connected systems hard to scale. With a shared standard, a team builds one server for a service and any app that speaks MCP can use it.

So when someone says “I added the GitHub MCP server to Claude,” they mean Claude can now read repositories and work with issues and pull requests, because a server gave it that ability.

How do MCP servers work?

Three parts are involved, and the architecture overview names them clearly:

• The host: the AI application you use, such as Claude Code or Claude Desktop.
• The client: a component inside the host that keeps a dedicated connection to one server. The host creates one client for each server you connect.
• The server: the program that provides the context and tools.

Connect three servers and the app runs three clients, one per server. Each pair talks over the protocol, which is built on JSON-RPC 2.0, a plain message format.

Here is a typical request from start to finish:

1. You ask your AI app a question, say “which of my open pull requests are waiting on review?”
2. The app has already asked each connected server what it offers, using a request called tools/list. It knows a GitHub server has a tool for listing pull requests.
3. The model decides that tool fits your question, and the app sends the server a tools/call request with the right arguments.
4. The server does the work (in this case a call to GitHub) and sends the result back.
5. The model reads the result and writes your answer.

You never see the middle steps. You see the answer, and most apps show which tool was used and ask your approval before running it.

The three things a server can offer

The protocol defines three core primitives that servers expose:

• Tools: functions the AI can run to take an action, such as file operations, API calls and database queries.
• Resources: data that gives the AI context, such as file contents, database records and API responses.
• Prompts: reusable templates that help structure how the AI works with a task.

A database server shows how they fit together. It can expose a tool for running queries, a resource that holds the database schema and a prompt with example questions. Most servers you will meet lean on tools, and that is where to focus when you are starting out. If you later want to write one, our guide to building your first MCP server walks through it.

Local vs remote MCP servers

“MCP server” means the program that serves context, wherever it runs. The docs describe two common setups, and the difference matters for setup and for security.

A local server is started by your AI app as a process on your computer, so it can reach what your computer can reach. That is the point of the Filesystem server, which reads, writes and searches files in folders you allow. A remote server lives with the service. The official docs use Sentry’s server as the example of one that runs on the vendor’s platform, and the protocol recommends OAuth for getting authentication tokens.

Some servers come both ways. GitHub’s server has a remote version and a local version that runs from a Docker image, and it supports both transports.

MCP server examples from the Klyvoo directory

Reading about tools is abstract. These five show the range, and each links to its full listing with install steps and price.

• Filesystem MCP: reads, writes, searches and organizes files in folders you allow. It is an official, local server and it is free and open source. Installing takes one npx command plus a list of folders it may touch.
• GitHub MCP Server: reads repositories, manages issues and pull requests and checks GitHub Actions runs. It is official, free and needs a GitHub account. Sign-in is OAuth or a personal access token.
• Playwright MCP: drives a real browser to click, type, navigate and read pages. It is an official Microsoft server that is free and open source, with no sign-in needed. Start it with npx @playwright/mcp@latest.
• Supabase MCP: manages Supabase projects, runs SQL, applies migrations and deploys edge functions.
• Context7: gives your assistant up-to-date, version-specific library docs and code examples.

The directory groups its 100 servers into categories, so you can browse by job: developer tools, browser and web, data and databases and files and storage are the ones most people start with.

What can you do with MCP servers?

The official introduction lists a few uses. Agents can reach your Google Calendar and Notion and act as a more personal assistant. Claude Code can build a web app from a Figma design. Company chatbots can connect to several databases and let people analyze data by chatting.

In daily work, the pattern is usually one of these:

• Coding: a coding agent reads your repo, searches docs and runs tests. Our best AI coding agents page shows the tools that support this, including Claude Code and Cursor.
• Research: a browser server lets an assistant open pages and read them instead of guessing.
• Operations: a project, chat or CRM server lets an assistant look things up and draft updates in the tools your team already uses.

If you use Claude, the step-by-step setup is in our guide to using MCP servers in Claude.

Are MCP servers safe?

They can be, if you treat them like any software you install. A local server runs with the same privileges as the app that starts it. The official MCP security guidance says the risks of local servers include arbitrary code execution, no visibility into what commands run and data loss. A server you add is code you are trusting.

A short checklist covers most of it:

1. Prefer official servers from the service’s own maker, and read the listing before you install. Our directory marks who made each one.
2. Read the command. Local servers are started by a command in your config. Know what it runs before you approve it.
3. Limit access. Give a file server only the folders it needs. For services with sign-in, use the narrowest permissions the server allows. The security docs recommend least-privilege scopes for the same reason: a stolen broad token can reach far more than a narrow one.
4. Keep write actions on approval. Reading is low risk. Sending, deleting and paying deserve a human click.
5. Remove what you stop using. Every server you connect adds tools to the assistant’s context, and some servers add a lot.

What MCP servers can’t do

They are connectors, not magic. A server does not make a model smarter or more accurate. It only gives the model more to work with, and a model can still call the wrong tool or misread a result.

The protocol also leaves the AI side to the app. The official docs say MCP focuses on the protocol for context exchange and does not dictate how AI applications use language models or manage the context they receive. Two apps can connect to the same server and behave differently.

Finally, a free server is not always a free service. The GitHub server costs nothing, but you need a GitHub account. Check the service’s own pricing before you build a workflow on it.

How to pick your first MCP server

Start with the job, not the server. Ask what you do every week that involves copying information between your AI app and another tool. Then:

• If you code, begin with GitHub MCP Server or Context7. Both fit a coding agent from the first day.
• If you work with files, begin with Filesystem MCP, limited to one project folder.
• If you research the web, try Playwright MCP.
• If you live in a database, look at the data and databases category.

Install one, use it for a week and add a second only when you feel the gap. Three well-chosen servers beat fifteen you never touch.

FAQ

What is an MCP server in simple terms?

It is a small program that lets an AI app use one outside system, such as your files or GitHub. The app asks the server what it can do and then uses those tools when your request needs them.

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

The client is the part inside your AI app that keeps one connection to a server. The server is the program on the other end that provides tools, data and prompts. An app that connects to three servers runs three clients.

Do I need to know how to code to use MCP servers?

Not always. Many apps let you add a remote server by signing in, and local servers usually need one line of setup in a config file or a command. Building your own server does need code.

Are MCP servers free?

Many are. The Filesystem, Playwright and GitHub servers listed above are free, but a server only connects you to a service, and that service may have its own plans and limits. Each of our server listings shows the price.

Which apps support MCP?

The official docs name Claude, ChatGPT, Visual Studio Code and Cursor among many others. Support and setup steps differ from app to app, so check your app’s own docs.

Is MCP the same as an API?

No. An API is one service’s own interface. MCP is a shared standard that sits on top and lets any compatible AI app discover and use a server’s tools the same way, without custom code for each service.

Ready to try one? Browse the Klyvoo MCP servers directory to compare servers by job, sign-in method and price.