# What is an MCP server? A plain-English guide (2026)

Canonical: https://theagenticweekly.com/what-is/mcp-server · Last verified: 2026-10-25 · Published: 2026-10-25 · Author: Agentic Weekly (operated independently from Texas, US) · Evidence: Evidence: explanations and examples are Reported from the official MCP docs and each vendor's own pages, read on 2026-10-25. The only thing we ran is FastMCP (a smoke test with no LLM); we did not run the example servers listed here.

> An MCP server is a program that gives AI applications access to data and actions through the Model Context Protocol (MCP), an open-source standard for connecting AI applications to external systems. The AI application (the host) opens one connection per server, and the server offers tools, resources and prompts. Servers run locally over stdio or remotely over HTTP. Examples include GitHub's, Playwright's and Figma's official servers. We smoke-tested one MCP framework, FastMCP, and everything else here is reported from official docs.

## MCP server meaning in one paragraph

MCP stands for Model Context Protocol. The official docs call it "an open-source standard for connecting AI applications to external systems" and compare it to a USB-C port: one standard way to plug AI applications into data sources (such as local files or databases), tools (such as search engines or calculators) and workflows (such as specialised prompts). An MCP server is the program on the other end of that connection: it "provides context to MCP clients". If you searched for what an MCP server means, that is it: a connector that lets an AI app read data or take actions in some other system.

The docs list examples of what this enables: agents using your Google Calendar and Notion, Claude Code generating a web app from a Figma design, and enterprise chatbots connecting to several databases.

## How an MCP server works: hosts, clients and servers

MCP uses a client-server design with three roles, defined in the architecture docs:

- **MCP host**: the AI application, such as Claude Code, Claude Desktop or Visual Studio Code. It manages one or more clients.
- **MCP client**: a component inside the host that keeps a dedicated connection to one server.
- **MCP server**: the program that provides context and actions to the client.

So the question "MCP server vs client" has a simple answer: the client lives inside the AI application and asks; the server provides. A host creates one client for each server it connects to. The docs add that "MCP server" refers to the program, wherever it runs.

The protocol has two layers. The data layer uses JSON-RPC 2.0 for the messages and defines the core features. The transport layer carries those messages, and it handles authentication for remote servers.

## What an MCP server can offer: tools, resources and prompts

Table: The three server primitives in the MCP architecture docs

| Primitive | What it is | Example from the docs |
|---|---|---|
| Tools | Executable functions the AI application can invoke to perform actions | File operations, API calls, database queries |
| Resources | Data sources that give the AI application context | File contents, database records, API responses |
| Prompts | Reusable templates that structure interactions with a model | System prompts, few-shot examples |

A client can ask a server what it offers (for example with `tools/list`) and then call a tool (`tools/call`). The docs give a database server as a concrete case: it can expose tools for querying, a resource holding the schema, and a prompt with few-shot examples. MCP only defines how context is exchanged; it does not dictate how the AI application uses models or manages that context.

## Local vs remote MCP servers

Table: The two transports in the MCP docs

| Kind | Transport | How it is described |
|---|---|---|
| Local | Stdio | Runs as a process on your machine and talks over standard input and output; typically serves a single client; no network overhead |
| Remote | Streamable HTTP | Uses HTTP POST with optional Server-Sent Events; can serve many clients; supports bearer tokens, API keys or custom headers, and MCP recommends OAuth for tokens |

The docs use the official Sentry MCP server as a remote example (it runs on Sentry's platform) and the filesystem server as a local one.

## Examples of MCP servers (with official pages)

This is our selection of widely referenced servers from well-known vendors, not a popularity ranking. We have not run any of these; descriptions come from each project's own README or docs, read on 2026-10-25, and are Reported.

Table: Example MCP servers and where to find them

| Server | What its own docs say it does | Run it | Official page |
|---|---|---|---|
| GitHub MCP Server | Connects AI tools directly to GitHub: read repositories and code files, manage issues and pull requests, analyse code, automate workflows. MIT license on GitHub | Hosted remote server, or a local version | [github/github-mcp-server](https://github.com/github/github-mcp-server) |
| Playwright MCP | Browser automation with Playwright: lets LLMs interact with web pages through structured accessibility snapshots instead of screenshots. Apache-2.0 license; needs Node.js 18 or newer | Local, started with `npx @playwright/mcp@latest` | [microsoft/playwright-mcp](https://github.com/microsoft/playwright-mcp) |
| Figma MCP server | Gives AI agents design context to generate code from Figma files and, per Figma, lets agents write content back to the canvas | Remote server at `https://mcp.figma.com/mcp` (recommended) or desktop server | [Figma developer docs](https://developers.figma.com/docs/figma-mcp-server/) |
| Reference servers | Fetch, Filesystem, Git, Memory, Sequential Thinking, Time and an Everything test server, maintained by the MCP steering group as reference implementations, not production-ready solutions | Local, with `npx` or `uvx` | [modelcontextprotocol/servers](https://github.com/modelcontextprotocol/servers) |

To browse more, the project points to the [MCP Registry](https://registry.modelcontextprotocol.io/), where published servers are listed. Its own servers repository says it is dedicated to the small set of reference servers, and tells you to use the registry to find others.

## How to add an MCP server to a client such as Claude Code

The steps depend on the host. As one documented example, Figma's docs say to add its remote server to Claude Code with `claude mcp add --transport http figma https://mcp.figma.com/mcp`, start a new session, type `/mcp`, choose figma and select Authenticate. Claude Code's own docs describe MCP as an open standard for connecting AI tools to data sources and point to an MCP quickstart. Other hosts, including Cursor, VS Code and Codex, have their own settings screens, and Figma's docs show steps for each. We have not run these steps.

Only add servers you trust. Local servers run code on your machine, and the MCP docs publish security guidance, linked below.

## Building your own MCP server (and what we tested)

The MCP docs have a [build-a-server guide](https://modelcontextprotocol.io/docs/develop/build-server), and the reference servers repository lists SDKs for TypeScript, Python, Go, Java, Kotlin, C#, PHP, Ruby, Rust and Swift.

We ran one thing in this space. On 2026-10-25 we installed FastMCP 4.1.3 with uv, wrote a two-tool server and called it through an in-memory client, a stdio subprocess and HTTP on 127.0.0.1, and `fastmcp inspect` listed two tools. No LLM was involved, and we did not test OAuth or the security fixes in that release. Read the scope and notes on the [FastMCP page](https://theagenticweekly.com/tools/fastmcp). That is a smoke test of one framework, not an evaluation of MCP servers in general.

## Security: what to know before you connect a server

- The MCP project publishes [security best practices](https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices) covering attack vectors and good practice.
- The official servers repository warns that its servers are reference implementations meant as educational examples, not production-ready solutions, and that you should evaluate your own security requirements.
- Remote servers authenticate over HTTP; the MCP docs recommend OAuth for obtaining tokens.
- For our own caution: an MCP server can read data or take actions on your behalf, so give it only the access it needs.

## FAQ

### What is an MCP server?

An MCP server is a program that provides context and actions to AI applications through the Model Context Protocol, an open-source standard for connecting AI applications to external systems. It can expose tools, resources and prompts, and can run locally or remotely.

### What does MCP stand for?

MCP stands for Model Context Protocol. The official docs describe it as an open-source standard for connecting AI applications to external systems, comparable to a USB-C port for AI applications.

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

The MCP client is a component inside the AI application (the host) that keeps a dedicated connection to one server. The MCP server is the program that provides context and actions. A host creates one client for each server it connects to.

### What can an MCP server do?

Per the MCP architecture docs, a server can expose tools (executable functions such as API calls or database queries), resources (data such as file contents or records) and prompts (reusable templates). Examples include GitHub's server for repositories and issues, Playwright's for browser automation and Figma's for design context.

### Are MCP servers local or remote?

Both exist. Local servers run on your machine over the stdio transport and typically serve one client. Remote servers use the Streamable HTTP transport and can serve many clients, with OAuth recommended for authentication.

### How do I add an MCP server to Claude Code?

It depends on the server. As one documented example, Figma's docs give `claude mcp add --transport http figma https://mcp.figma.com/mcp`, then `/mcp` in a new session to authenticate. We have not run these steps.

### Are MCP servers safe?

That depends on the server and how you configure it. The MCP project publishes security best practices, and its reference servers are described as examples, not production-ready solutions. Connect only servers you trust and limit their access.

### Have you tested MCP servers?

We smoke-tested FastMCP 4.1.3, a framework for building them, on 2026-10-25 with no LLM. We did not run the example servers on this page; their descriptions are Reported from official pages.

## Sources

- [What is the Model Context Protocol (MCP)? (official docs)](https://modelcontextprotocol.io/docs/getting-started/intro) (fetched 2026-10-25)
- [MCP architecture overview (official docs)](https://modelcontextprotocol.io/docs/learn/architecture) (fetched 2026-10-25)
- [MCP reference servers (GitHub)](https://github.com/modelcontextprotocol/servers) (fetched 2026-10-25)
- [MCP Registry](https://registry.modelcontextprotocol.io/) (fetched 2026-10-25)
- [GitHub MCP Server (GitHub)](https://github.com/github/github-mcp-server) (fetched 2026-10-25)
- [Playwright MCP (GitHub)](https://github.com/microsoft/playwright-mcp) (fetched 2026-10-25)
- [Figma MCP server (Figma developer docs)](https://developers.figma.com/docs/figma-mcp-server/) (fetched 2026-10-25)
- [Figma remote server setup](https://developers.figma.com/docs/figma-mcp-server/remote-server-installation/) (fetched 2026-10-25)
- [Claude Code overview (MCP section)](https://code.claude.com/docs/en/overview) (fetched 2026-10-25)

## Related

- [FastMCP: our smoke test](https://theagenticweekly.com/tools/fastmcp)
- [Best AI agent tools](https://theagenticweekly.com/best/agent-building)
- [Claude Code](https://theagenticweekly.com/tools/claude-code)
- [How we pick and test](https://theagenticweekly.com/methodology)
- [Guide: Playwright MCP server](https://theagenticweekly.com/guides/playwright-mcp-server)
- [Guide: Git MCP server](https://theagenticweekly.com/guides/git-mcp-server)
- [Guide: GitHub MCP server](https://theagenticweekly.com/guides/github-mcp-server)
- [Guide: add an MCP server to Claude Code](https://theagenticweekly.com/guides/add-mcp-server-to-claude-code)
- [Guide: build an MCP server with FastMCP](https://theagenticweekly.com/guides/build-an-mcp-server-with-fastmcp)

## Changelog

- 2026-10-25: First published from the official MCP docs and vendor pages read on this date.

## How to cite

Agentic Weekly, "What is an MCP server? A plain-English guide (2026)", last updated 2026-10-25, https://theagenticweekly.com/what-is/mcp-server