// src/index.ts
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-server", version: "1.0.0" });

// Define a tool
server.tool("get_weather", { city: z.string() }, async ({ city }) => ({
  content: [{ type: "text", text: `Weather in ${city}: 72°F, sunny` }],
}));

// Define a resource
server.resource("config", "config://app", async () => ({
  contents: [{ uri: "config://app", text: JSON.stringify({ env: "production" }) }],
}));

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

Build and Test

# Build
npm run build

# Test with MCP inspector
npx @modelcontextprotocol/inspector node dist/index.js

# Test tool call directly
echo '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_weather","arguments":{"city":"NYC"}}}' | node dist/index.js

Register in Claude Code Config

// ~/.claude/settings.json or project .claude/settings.json
{
  "mcpServers": {
    "my-server": {
      "command": "node",
      "args": ["/path/to/my-mcp-server/dist/index.js"],
      "env": {
        "API_KEY": "your-key"
      }
    }
  }
}

Python MCP Server

# Install SDK
pip install mcp

# Create server
cat > server.py << 'EOF'
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("my-server")

@mcp.tool()
def get_data(query: str) -> str:
    """Fetch data based on query."""
    return f"Results for: {query}"

@mcp.resource("config://app")
def get_config() -> str:
    """Application configuration."""
    return '{"env": "production"}'

if __name__ == "__main__":
    mcp.run()
EOF

# Run
python server.py

Tool Design Best Practices

Clear names: search_users not do_search. Verb + noun.
Typed inputs: Use Zod (TS) or type hints (Python) for all parameters.
Good descriptions: The AI reads these to decide when to use the tool.
Error handling: Return error messages, don't throw. The AI needs to understand what went wrong.
Idempotent reads: GET-like tools should be safe to call multiple times.
Confirm writes: Destructive tools should describe what they'll do before doing it.

Notes

MCP servers communicate over stdio (stdin/stdout). Don't print debug info to stdout.
Use stderr for logging: console.error("debug info").
Test with the inspector before connecting to an agent.
Keep tools focused — one tool per operation, not a mega-tool with mode switches.
Resources are for context the agent reads. Tools are for actions the agent takes.

mcp-builder

MCP Server Builder

Quickstart — Create an MCP Server

Server Structure (TypeScript)