MCP Just Got Its First Major Funding — Why This Protocol Is the USB-C of AI (And How to Build Your Own Server)

TL;DR: Manufact, a YC-backed startup, just raised $6.3M to build infrastructure around MCP (Model Context Protocol). This protocol is rapidly becoming the "USB-C for AI" — a universal standard that lets AI assistants like ChatGPT and Claude connect to external tools and data. Here's what MCP is, why it matters, and how I built my own MCP server in a weekend.

---

What Just Happened

This week, Manufact announced a $6.3 million funding round to build the infrastructure layer for MCP. Backed by Y Combinator, Manufact is betting that MCP will become the universal connector between AI models and the real world — the same way USB-C became the universal connector for physical devices.

This isn't just another startup funding round. It's a signal that the AI ecosystem is maturing beyond isolated chatbots into a connected, interoperable network of tools and services.

What Is MCP, Exactly?

MCP (Model Context Protocol) is an open protocol that standardizes how AI assistants communicate with external tools, data sources, and services. Think of it as an API standard specifically designed for AI agents.

Before MCP, every AI tool integration was a custom build. If you wanted ChatGPT to query your database, you'd build one integration. If you wanted Claude to access the same data, you'd build another. Each AI provider had its own plugin system, its own authentication flow, its own everything.

MCP changes this by providing a universal protocol that any AI assistant can use to connect to any tool. Write your tool integration once using MCP, and it works with every MCP-compatible AI client.

The Analogy: USB-C for AI

The USB-C analogy is perfect. Before USB-C, we had a drawer full of different cables — Lightning, Micro-USB, Mini-USB, proprietary connectors. Each device needed its own cable. USB-C standardized everything: one connector, one cable, works everywhere.

That's exactly what MCP does for AI integrations. Instead of building custom integrations for each AI assistant, you build one MCP server that any MCP-compatible client can use.

Why Both ChatGPT and Claude Use MCP

Here's what makes this interesting: Anthropic created MCP and open-sourced it, and OpenAI has adopted it for ChatGPT. When the two biggest AI companies agree on a standard, you pay attention.

• ChatGPT now supports MCP servers, letting you connect your AI assistant to custom tools and data sources
• Claude has had MCP support built-in since the beginning — it was designed from the ground up to be an agent that uses tools
• Other AI tools are rapidly adopting MCP: Cursor, Windsurf, VS Code Copilot, and more

This convergence means developers can build tool integrations once and reach users across all major AI platforms.

The $6.3M Question: What Is Manufact Building?

Manufact is positioning itself as the "Stripe for MCP" — the infrastructure company that makes it easy for developers to build, deploy, and monetize MCP servers. With $6.3M in funding, they're betting that:

1. Every company will need MCP servers to expose their data and tools to AI assistants
2. Developers will need tools to build, test, and deploy these servers
3. A marketplace will emerge for discovering and connecting to MCP servers

This is the "NPM of AI" moment — when a standard protocol creates an entire ecosystem of packages, tools, and marketplaces around it.

My Experience: Building taiwan-flight-mcp

To understand MCP's potential, I built my own MCP server: taiwan-flight-mcp, a published NPM package that lets AI assistants search for Taiwan flight information.

Why I Built It

I travel between Taiwan and other countries frequently, and I got tired of manually checking flight information. I wanted my AI assistant to be able to search flights directly. So I built an MCP server that:

• Queries Taiwan's Civil Aviation Administration data
• Returns flight schedules, statuses, and delays
• Works with any MCP-compatible AI client

How It Works

The MCP server exposes three core capabilities (called "tools" in MCP terminology):

1. Search flights — Find flights by route and date
2. Get flight status — Real-time delay and gate information
3. Get airport info — Current conditions and facilities

When I ask Claude or ChatGPT "Are there any delays on flights from Taipei to Tokyo today?", the AI assistant calls my MCP server, gets the data, and gives me an answer — all without me opening a browser.

Building It Was Surprisingly Simple

Here's the basic structure of an MCP server:

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

const server = new McpServer({
  name: "taiwan-flight-mcp",
  version: "1.0.0",
});

server.tool("search-flights", {
  departure: { type: "string", description: "\"Departure airport code\" },"
  arrival: { type: "string", description: "\"Arrival airport code\" },"
  date: { type: "string", description: "\"Flight date (YYYY-MM-DD)\" },"
}, async (params) => {
  // Call the aviation API and return results
  const flights = await searchFlights(params);
  return { content: [{ type: "text", text: JSON.stringify(flights) }] };
});

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

That's it. The MCP SDK handles all the protocol details — you just define your tools and their handlers.

Publishing to NPM

I published taiwan-flight-mcp to NPM, which means anyone can install it with:

npm install -g taiwan-flight-mcp

And configure it in their AI assistant's MCP settings. No custom build needed, no server to maintain — it's just an NPM package.

How to Build Your Own MCP Server

Want to build your own? Here's a quick guide:

1. Set Up Your Project

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

2. Create Your Server

Create an index.ts file and define your tools using the pattern above. Each tool is a function that takes structured input and returns structured output.

3. Define Your Tools

Think about what data or capabilities you want to expose:

• Database queries
• API access
• File system operations
• Custom business logic

Each becomes a server.tool() call.

4. Test Locally

Most MCP clients (like Claude Desktop or Cursor) let you configure local MCP servers for testing. Add your server to the client's config, and start testing.

5. Publish

Package it up and publish to NPM. Now anyone can use your MCP server with their AI assistant.

Why This Is the "NPM of AI" Moment

Remember when NPM launched? It was just a package manager. But it created an ecosystem:

• Package developers could share their code
• App developers could find and use packages
• Companies built businesses around the ecosystem (npm Inc, Sonatype, Snyk)

MCP is creating the same dynamic for AI:

• MCP server developers share their tool integrations
• AI users discover and connect to useful tools
• Companies like Manufact build infrastructure and marketplaces

The total addressable market is enormous. Every company that wants their product to be accessible to AI assistants will need an MCP server. Every developer who wants their AI assistant to do more will use MCP tools.

What This Means for Developers

If you're a developer today, here's my advice:

1. Learn MCP now. It's early enough that being an expert gives you a real advantage
2. Build something. The best way to learn is to build your own MCP server. Start with something you use daily
3. Publish it. NPM makes distribution trivial. Get your name out there
4. Think protocol-first. Build tools that follow MCP standards, not platform-specific integrations

The developers who built NPM packages in 2010 had a huge head start. The developers building MCP servers in 2025 will have the same advantage.

Try It Yourself

If you want to see MCP in action, check out:

• taiwan-flight-mcp — My Taiwan flight search MCP server. Install it and connect it to your AI assistant
• Open Code Review — Another open-source project that's exploring MCP integration for AI-powered code quality checks in CI/CD pipelines
• Model Context Protocol — The official MCP documentation and specification

The future of AI is not one monolithic assistant that does everything. It's a network of specialized tools connected through a universal protocol. MCP is that protocol, and the time to start building is now.

---

What MCP server would you build? I'd love to hear about your ideas. Find me on GitHub or try out taiwan-flight-mcp and let me know what you think!