Creating Use Case-Specific MCP Servers with Postman

Today's AI agents rely on external data—such as real-time crypto prices, breaking news, or space updates. But there’s a problem: typical off-the-shelf MCP (Multi-Cloud Platform) servers are often heavy, difficult to inspect, and not built for focused tasks. On the other hand, building a server from scratch takes a lot of time and effort for developers.

A Simple Idea

Josh Dzielak from Postman has introduced a practical approach. Imagine selecting just the API endpoints you need from Postman’s public API network. From there, an MCP server is automatically created to match your requirements. It’s like having a server tailored specifically for your AI agent. This method balances:

Security: You control the code and know exactly what’s inside.
Performance: No extra components slowing things down.
Simplicity: Fewer tools mean easier setup and fewer complications.

Getting Started with TypeScript

If you want to build your own focused MCP server, here is a guide using Node.js and TypeScript:

1. Project Structure

my-time-agent/
├── tools/
│   ├── coingecko/
│   ├── nyt/
│   └── nasa/
├── .env
├── mcp-server.ts
└── package.json

2. Main Server Code (mcp-server.ts)

import { createServer } from 'mcp-server';
import { coingeckoTool } from './tools/coingecko/tool';
import { nytTool } from './tools/nyt/tool';
import { nasaTool } from './tools/nasa/tool';

createServer({
  tools: [coingeckoTool, nytTool, nasaTool],
  model: 'gpt-4', // or choose your preferred model
  io: 'stdio', // or use HTTP
});

3. Example Tool (tools/coingecko/tool.ts)

import { defineTool } from 'mcp-server';
import axios from 'axios';

export const coingeckoTool = defineTool({
  name: 'coin_price_history',
  description: 'Get historical crypto prices by date',
  inputSchema: {
    coinId: 'string',
    date: 'string', // format: dd-mm-yyyy
  },
  execute: async ({ coinId, date }) => {
    const res = await axios.get(`https://api.coingecko.com/api/v3/coins/${coinId}/history`, {
      params: { date },
    });
    return res.data.market_data.current_price;
  },
});

4. Running the Server

npm install
npm run build
node dist/mcp-server.js

Once the server is running, you can connect to it using an MCP client like Cursor or Postman.

Key Points

Smaller is Better: A server with only what you need runs faster and is easier to manage.
Custom Yet Familiar: MCP provides a common structure, but you can adapt it to your needs.
Postman’s Help: Their generator removes the need for complex initial setup.

Josh’s toolchain also encourages careful testing with Cursor, keeping track of API versions, and easy setup of environment variables.

Acknowledgements

This guide is based on Josh Dzielak’s talk at the MCP Summit, where he explained how Postman can help build focused MCP servers. Watch the demonstration here: Generating Use Case-Specific MCP Servers.

Thanks to the Anthropic team and the broader MCP developer community for their ongoing contributions to building better tools for AI agents.