How to Install and Configure Your First MCP Server

Setting up your first MCP server comes down to three real steps: installing an official SDK, writing a minimal server that registers at least one tool, and pointing an actual MCP client at it. Here's each step with working code, not placeholder pseudocode.

Step 1: Install an Official SDK

Anthropic maintains official SDKs for both major languages:

# TypeScript/JavaScript
npm install @modelcontextprotocol/sdk

# Python
pip install mcp

Step 2: Write a Minimal Server

Here's a genuinely minimal, working stdio-based server using the TypeScript SDK, registering one 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: "my-first-server", version: "1.0.0" });

server.registerTool(
  "echo",
  {
    description: "Echoes back the input string",
    inputSchema: { message: z.string() }
  },
  async ({ message }) => ({
    content: [{ type: "text", text: message }]
  })
);

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

This is a complete, runnable server — not a fragment. It exposes exactly one tool (echo) over the stdio transport, which is what most local MCP clients (Claude Desktop, most CLI-based agents) expect for locally-run servers.

Step 3: Point Claude Desktop at It

Claude Desktop reads its MCP server list from a real config file — claude_desktop_config.json, located under your OS's application support directory. Add your server:

{
  "mcpServers": {
    "my-first-server": {
      "command": "node",
      "args": ["/absolute/path/to/your/server.js"]
    }
  }
}

Restart Claude Desktop after saving. If the server starts correctly, its tools appear in the client's tool list — you can verify this directly by asking Claude something that would require the echo tool.

Step 4: Test With the Official Inspector Instead of Guessing

Before wiring a new server into a full AI client, Anthropic's own MCP Inspector tool lets you exercise it directly:

npx @modelcontextprotocol/inspector node /absolute/path/to/your/server.js

This opens a local web UI where you can call tools/list, invoke tools/call with real arguments, and see the raw JSON-RPC messages going back and forth — far faster for debugging a broken tool schema than trying to reproduce the issue through a full chat interface.

Common First-Run Problems

• Relative paths in config: use the full absolute path to your server file — a relative path may not resolve correctly depending on the client's working directory.
• Wrong transport: a server built for stdio won't work if a client expects HTTP/SSE, and vice versa — check what transport your target client actually supports before writing the server.
• Silent startup failures: stdio servers that crash on launch often fail silently from the client's perspective — run the server directly from a terminal first to see any startup errors before wiring it into a client config.