MCPservers,zerotoone

What an MCP server actually is#

A Model Context Protocol server is a small process that exposes tools, resources, and prompts to an AI client. Think of an MCP server as a USB-C port for AI: a single, well-defined way to plug capabilities into Claude.

You write the server. You describe what it can do. Claude, or any MCP-aware client, queries it and uses the tools you exposed.

The minimum viable setup#

Install the SDK:

npm install @modelcontextprotocol/sdk

Then a server file with one tool:

import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

const server = new Server({ name: "my-first-mcp", version: "0.1.0" });

server.setRequestHandler("tools/list", async () => ({
  tools: [{
    name: "ping",
    description: "Health check",
    inputSchema: { type: "object", properties: {} },
  }],
}));

server.setRequestHandler("tools/call", async (req) => {
  if (req.params.name === "ping") {
    return { content: [{ type: "text", text: "pong" }] };
  }
});

await server.connect(new StdioServerTransport());

Wire it up#

Register the server in your Claude Code settings under mcpServers. Restart Claude. Call ping. You'll see pong.

That's the loop. Everything else — auth, real tools, resources — is just expanding what tools/list returns and how tools/call responds.

What to build first#

Your first useful MCP server is almost always a wrapper around a CLI you already use. Wrap gh, vercel, or your internal scripts. The friction of "agent runs my tool" drops to zero. That's where the productivity unlock starts.