What is MCP?

The problem MCP solves

AI agents are great at reasoning, writing, and working with code. But on their own, they can’t check your Slack messages, query a database, create a Jira ticket, or search the web. They’re stuck in a sandbox.

Before MCP, every integration was custom. Each agent had its own plugin format, its own API, and its own way of connecting to tools. If you built a Slack integration for one agent, you had to rebuild it for another.

MCP standardizes this. Build one MCP server, and it works with every agent that supports the protocol.

How MCP works

MCP follows a client-server architecture. The AI agent is the client. The external service is the server. They communicate over a standardized protocol.

When an agent connects to an MCP server, it asks: “What can you do?” The server responds with a list of tools, each with a name, description, and input schema. The agent can then decide when and how to use them.

The three primitives

MCP servers can expose three types of capabilities:

Most MCP servers focus on tools — they give the agent new abilities. Resources and prompts are useful but less commonly implemented.

What an MCP server looks like

Here’s a minimal MCP server in TypeScript that provides a single tool:

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";

const server = new McpServer({
  name: "weather",
  version: "1.0.0",
});

server.tool("get_weather",
  { city: z.string() },
  async ({ city }) => {
    const data = await fetch(
      `https://api.weather.com/v1/${city}`
    ).then(r => r.json());
    return {
      content: [{
        type: "text",
        text: `${city}: ${data.temp}°F, ${data.condition}`,
      }],
    };
  }
);

const transport = new StdioServerTransport();
await server.connect(transport);

That’s it. The agent can now ask for the weather in any city. It discovers the get_weather tool, sees it takes a city string, and calls it when relevant.

Connecting MCP servers to agents

Each agent has its own way of registering MCP servers. Here are the most common:

{ "mcpServers": { "my-server": { "command": "npx", "args": ["-y", "my-mcp-server"] } } }

Transport options

MCP servers communicate with agents through one of two transports:

Which agents support MCP?

MCP adoption is growing rapidly. Agents with MCP support include:

MCP vs Skills

MCP and Skills solve different problems. MCP gives agents new abilities (access to external systems). Skills give agents domain knowledge (how to do specific tasks well).

For a deeper comparison, see Skills vs MCP Servers.

Getting started