What Is an MCP Server? A Plain-English Explanation
Updated 2026-11-18
An MCP server is a program that gives an AI application access to tools, data or prompt templates through the Model Context Protocol (MCP). The AI app, called the host, connects to the server, asks what it offers, and calls those tools when a task needs them. MCP is an open-source standard, so one server works with any client that supports the protocol.
The short version
The official MCP documentation describes the protocol as a USB-C port for AI applications. Without a standard, every AI app needs custom code for every service. With MCP, a service publishes one server and any compatible app can plug into it.
Examples of what a server can do: search a code repository, read a Notion page, create a payment link, query a database, or deploy a Cloudflare Worker. The server exposes these as named functions that the model can call.
How it works: host, client, server
MCP uses a client-server design with three parts:
- Host: the AI application, such as Claude Code, Cursor, VS Code or ChatGPT.
- Client: a component inside the host that keeps one dedicated connection to one server.
- Server: the program that provides context and actions.
If you connect three servers, the host runs three clients. Each one talks to its own server. The model sees the combined list of tools and decides which to call.
What a server can expose
MCP defines three core things a server can offer:
| Primitive | What it is | Example |
|---|---|---|
| Tools | Functions the AI can invoke to take an action | Create an issue, run a query, send a message |
| Resources | Data that gives the AI context | File contents, database records, API responses |
| Prompts | Reusable templates that structure an interaction | A code-review prompt with few-shot examples |
Most servers you will find in a directory are mainly tool servers. Each tool has a name, a description and an input schema. The model reads the description to decide when to use the tool.
{
"name": "get_weather",
"description": "Get current weather information for a location",
"inputSchema": {
"type": "object",
"properties": {
"location": { "type": "string", "description": "City name or zip code" }
},
"required": ["location"]
}
}
This is a tool definition as shown in the MCP specification. The client lists tools with a tools/list request and runs one with tools/call.
Local and remote servers
The word "server" does not mean the program runs somewhere else. MCP supports two transports:
- stdio: the client starts the server as a process on your machine and talks to it through standard input and output. This is a local server. It has the same access to your machine that you give it.
- Streamable HTTP: the server runs at a URL and the client sends HTTP requests. This is a remote server. The MCP docs say it supports bearer tokens, API keys and custom headers, and recommend OAuth to obtain tokens.
The Sentry MCP server is the official example of a remote server. The reference filesystem server is the usual example of a local one.
Remote servers are easier to use because there is nothing to install. They are also the only kind that web-based clients such as ChatGPT and claude.ai can reach.
How many MCP servers exist?
The official MCP Registry is the public metadata catalog for servers. According to Agents Index probe data, Nov 2026, the registry held 65,640 servers (latest versions), and 44,880 of them expose a remote endpoint. We probe those endpoints daily. Of the remote endpoints, 23,704 answered without sign-in, 13,608 were online but required sign-in (OAuth or API key), and 7,189 were unreachable. The full breakdown is in the State of MCP report.
The takeaway: a listing in a registry does not mean the server works. Check that an endpoint is reachable before you spend time configuring it.
How to try one
The fastest way to see an MCP server work is to add a remote server to a client that supports it. In Claude Code, for example, one command connects Notion:
claude mcp add --transport http notion https://mcp.notion.com/mcp
Then run /mcp inside Claude Code to sign in. Ask the assistant to find a page and watch it call the server's tools. The step-by-step version is in how to add an MCP server to Claude Code. You can browse servers that work with Claude Code.
What to check before you connect
- Who publishes it. Servers in the official registry use namespaces tied to a GitHub account or a domain, which shows who registered the name.
- What it can do. Read the tool list. A server with write or delete tools needs more trust than a read-only one.
- Where it runs. A local server runs code on your machine with your permissions. A remote server sends your requests to someone else's infrastructure.
The MCP server security checklist goes through these points in detail.
Where to find servers
Start at the Agents Index home page. Servers published to the official registry appear in our official registry listing.
Sources
- What is the Model Context Protocol (MCP)? - modelcontextprotocol.io
- MCP architecture overview - modelcontextprotocol.io
- MCP specification: Tools - modelcontextprotocol.io
- Connect Claude Code to tools via MCP - code.claude.com
- The MCP Registry - modelcontextprotocol.io
FAQ
What does MCP stand for?
MCP stands for Model Context Protocol, an open-source standard for connecting AI applications to external systems such as data sources, tools and workflows.
Is an MCP server the same as an API?
No. An API is built for developers to call from code. An MCP server wraps capabilities in a standard format so an AI app can discover and call them without custom integration code.
What is the difference between a local and a remote MCP server?
A local server runs on your machine and talks to the client over stdio. A remote server runs on the provider's infrastructure and is reached over HTTP at a URL, usually with OAuth or an API key.
Do I need to code to use an MCP server?
No. Most clients let you add a server by pasting a URL or a short config entry. You only write code if you are building your own server.