Deutsch
Agent protocols

MCP server

An MCP server is a program that exposes capabilities such as functions, data and prompt templates to AI applications through the Model Context Protocol (MCP). “Server” names a role in the protocol, not a specific machine. It can run on a laptop next to the AI application or as a service on the web.

At a glance

Publisher
Anthropic, since 2025-12-09 at the Agentic AI Foundation (Linux Foundation)
Connects
AI applications with servers that provide tools, resources and prompts
Status
current MCP specification 2026-07-28
Transport
stdio (local servers), Streamable HTTP (remote servers)
Specification
MCP Specification

Last checked:

How does an MCP server work?

The AI application, called the host in MCP, opens one client connection per server. Through that connection it asks the server what it offers and requests what the model needs. The MCP documentation describes three building blocks:

  • Tools are functions the model can call, such as a database query, an API request or a file edit.
  • Resources are read-only data sources, such as file contents or database schemas. Each has a URI.
  • Prompts are reusable instruction templates that a user invokes explicitly.

Clients discover these through list methods like tools/list and run a tool with tools/call. Local servers typically talk to the host over stdio. Remote servers use Streamable HTTP and can serve many clients, with OAuth as the recommended way to authenticate.

MCP: host, clients and servers, plus WebMCPAn AI application acting as host starts one MCP client per server. Each server offers tools, resources and prompts. Local servers connect via stdio, remote servers via Streamable HTTP. Below, WebMCP: a web page registers tools in JavaScript, and an agent in the browser calls them.HOSTMCP SERVERSHostAI application, e.g. a chat appMCP client 1stdioServer 1 · fileslocalToolsResourcesPromptsMCP client 2stdioServer 2 · databaselocalToolsResourcesPromptsMCP client 3Streamable HTTPServer 3 · external APIremoteToolsResourcesPromptsWEBMCP · DRAFT OF THE W3C WEB MACHINE LEARNING COMMUNITY GROUPAgent in the browsere.g. the browser’s agentcalls a toolWeb pageregisters JavaScript tools that run in the browserSchematic. As of MCP specification 2026-07-28 and WebMCP draft of Oct 2, 2026.
According to the MCP specification, a host starts one client per server. Servers offer tools, resources and prompts and run locally (stdio) or remotely (Streamable HTTP). WebMCP is a draft by a W3C community group, not a W3C standard. Here the web page provides its tools inside the browser. Sources: Model Context Protocol: Specification 2026-07-28; Model Context Protocol: Architecture; Model Context Protocol: Transports; W3C Web Machine Learning Community Group: WebMCP (draft of Oct 2, 2026).

Example

A host calls a flight search tool on a travel server. The request is a JSON-RPC message, shortened here.

JSON
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "searchFlights",
    "arguments": { "origin": "NYC", "destination": "Barcelona", "date": "2026-06-15" }
  }
}

The server returns the flight options as structured content, and the host passes them to the language model. The documentation notes that tools may require user consent before they run, for example through an approval dialog.

Well-known MCP servers

The project maintains reference servers in the modelcontextprotocol/servers repository on GitHub, including a file system server. Many services publish their own servers for tasks like code management, messaging or calendars.

Common mistakes

  • Not checking inputs. According to the specification, servers must validate all tool inputs, enforce access controls, rate limit invocations and sanitize outputs.
  • Reporting invalid input as a protocol error. Input errors such as a wrong date format belong in the tool result as a tool execution error with isError: true, so the model can correct the request.
  • Unsuitable tool names. The specification calls for 1 to 128 characters from letters, digits, underscore, hyphen and dot, without spaces, unique within a server.

FAQ

What is the difference between MCP and an MCP server? MCP is the protocol. An MCP server is a program that speaks it and offers tools, resources or prompts.

What is the difference between an MCP server and an API? A web API is an interface for software in general. An MCP server wraps functions in a standard format that AI applications can discover and call. It can call regular APIs behind the scenes.

Sources

  1. Understanding MCP servers – modelcontextprotocol.io (spec version 2026-07-28)modelcontextprotocol.io
  2. Architecture overview – modelcontextprotocol.iomodelcontextprotocol.io
  3. MCP Reference Server Implementations – GitHubgithub.com
  4. WebMCP – Draft Community Group Report, W3C Web Machine Learning Community Groupwebmachinelearning.github.io
  5. Tools – MCP Specification (version 2026-07-28)modelcontextprotocol.io

Last updated: