MCP Power User — Custom Connections

When to build a custom MCP server

The 200+ community MCP servers cover the major tools. You need a custom server when:

• You use an internal tool that has no public MCP server
• You have a private API that Claude should be able to call
• You need Claude to interact with a system in a very specific, controlled way
• You want to expose internal company data to Claude without it leaving your network

Custom MCP servers are small Node.js programs. You don't need to be a senior engineer. If you can write basic JavaScript, you can build one in under an hour.

Build a custom MCP server from scratch

We'll build a server that gives Claude access to a simple REST API — in this case, a weather API and a custom note-taking function. This template works for any API.

Step 1: Initialize the project

mkdir my-mcp-server
cd my-mcp-server
npm init -y
npm install @modelcontextprotocol/sdk

Step 2: Create the server file

import { Server } from '@modelcontextprotocol/sdk/server/index.js';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
import { ListToolsRequestSchema, CallToolRequestSchema } from '@modelcontextprotocol/sdk/types.js';
import fs from 'fs';

// Create the MCP server
const server = new Server(
  { name: 'my-tools', version: '1.0.0' },
  { capabilities: { tools: {} } }
);

// Define available tools
server.setRequestHandler(ListToolsRequestSchema, async () => ({
  tools: [
    {
      name: 'save_note',
      description: 'Save a note to a local file for later reference',
      inputSchema: {
        type: 'object',
        properties: {
          title: { type: 'string', description: 'Note title' },
          content: { type: 'string', description: 'Note content' }
        },
        required: ['title', 'content']
      }
    },
    {
      name: 'get_weather',
      description: 'Get current weather for a city',
      inputSchema: {
        type: 'object',
        properties: {
          city: { type: 'string', description: 'City name' }
        },
        required: ['city']
      }
    }
  ]
}));

// Handle tool calls
server.setRequestHandler(CallToolRequestSchema, async (request) => {
  const { name, arguments: args } = request.params;

  if (name === 'save_note') {
    const filename = `./notes/${args.title.replace(/\s+/g,'-')}.md`;
    fs.mkdirSync('./notes', { recursive: true });
    fs.writeFileSync(filename, `# ${args.title}\n\n${args.content}`);
    return { content: [{ type: 'text', text: `Saved note to ${filename}` }] };
  }

  if (name === 'get_weather') {
    // Replace with your actual weather API key + endpoint
    const apiKey = process.env.WEATHER_API_KEY;
    const url = `https://api.openweathermap.org/data/2.5/weather?q=${args.city}&appid=${apiKey}&units=imperial`;
    const res = await fetch(url);
    const data = await res.json();
    const summary = `${data.name}: ${data.main.temp}°F, ${data.weather[0].description}`;
    return { content: [{ type: 'text', text: summary }] };
  }

  throw new Error(`Unknown tool: ${name}`);
});

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

Step 3: Update package.json for ES modules

{
  "name": "my-mcp-server",
  "version": "1.0.0",
  "type": "module",
  "main": "index.js",
  "scripts": {
    "start": "node index.js"
  }
}

Step 4: Register with Claude Desktop

{
  "mcpServers": {
    "my-tools": {
      "command": "node",
      "args": ["/absolute/path/to/my-mcp-server/index.js"],
      "env": {
        "WEATHER_API_KEY": "your_api_key_here"
      }
    }
  }
}

Restart Claude Desktop. You'll now see your custom tools in the tool list. Ask Claude: "Save a note titled 'Meeting Notes' with the content 'Discussed Q2 roadmap.'" Claude will call your save_note tool and create the file.

Security: what to know before you connect anything

• Filesystem access: Only grant access to specific directories, never your whole home folder
• API keys: Store in environment variables, never hardcode in scripts that get committed to git
• Network access: Know what external endpoints your MCP servers call
• Community servers: Check the GitHub repo, look at the source code before installing
• Work credentials: Be especially careful with servers that access work systems — check your company's AI policy first

For a home or personal development environment, the official Anthropic-maintained servers are safe. For corporate environments, discuss with your IT/security team before deploying MCP servers that access internal systems.

The future: agentic desktop AI

MCP is 6 months old as of early 2026. The ecosystem is expanding fast:

• Computer use: Claude can already control a computer — clicking, typing, navigating. Expect this to integrate more tightly with Desktop.
• Long-running agents: Background tasks that run for hours, check in with you, and complete complex multi-step work autonomously.
• Multi-agent coordination: Claude Desktop routing tasks to specialized sub-agents — one for research, one for writing, one for code review.
• Memory: Persistent knowledge that accumulates across sessions without you having to re-explain context.

The people who learn MCP now will have a significant advantage as these capabilities arrive. The concepts are the same: tools, context, agent loops. The scope just gets bigger.