Build your first MCP server in 5 minutes with mcp-creator

What is mcp-creator?

mcp-creator is an MCP server that builds other MCP servers. You install it, add it to your AI assistant, and describe what you want in plain English. It handles scaffolding, packaging, publishing, and GitHub setup.

No pyproject.toml to configure. No tsconfig.json to set up. No boilerplate server code to write. You talk, it builds.

Available in two flavors:

• mcp-creator-python for Python servers published to PyPI
• mcp-creator-typescript for TypeScript servers published to npm

Both are free and open source. Both are listed on MCP Marketplace.

Want the full deep-dive instead? See the Python MCP Creator guide or the TypeScript MCP Creator guide. This post is the fast version.

Prerequisites

You need one thing: an AI assistant that supports MCP. That means:

• Claude Desktop (Mac or Windows)
• Claude Code (terminal)
• Cursor, VS Code, or any other MCP-compatible editor

That's it. mcp-creator handles first-time setup for everything else (Python/Node.js, package managers, git, GitHub CLI, registry tokens). If something is missing, it tells you how to install it.

Step 1: Install mcp-creator

Pick your language. One command each.

Python:

pip install mcp-creator-python

TypeScript:

npx -y mcp-creator-typescript

Then add it to your AI assistant's config:

claude mcp add mcp-creator-python -- mcp-creator-python

For TypeScript, use npx -y mcp-creator-typescript as the command instead. See the TypeScript guide for the exact config snippet.

Restart your AI assistant. You're ready.

Step 2: Describe what you want to build

This is the whole point. Instead of writing code, you write English.

Open your AI assistant and say something like:

"Build me an MCP server that gets the current weather for any city using the OpenWeatherMap API"

mcp-creator takes it from there:

1. Checks your setup -- verifies Python/Node, package manager, git, and registry auth are all in place
2. Checks name availability -- confirms your package name isn't taken on PyPI or npm
3. Scaffolds the project -- generates the full directory structure, server entry point, tool definitions, service stubs, tests, README, and a marketplace submission checklist

You get a complete, runnable project in seconds. Not a starter template. A real project with proper packaging, entry points, and documentation.

Step 3: Fill in your logic and test locally

The generated project runs immediately with placeholder data. Your job is to fill in the real API calls.

For a weather server, that means opening services/weather_service.py (or the equivalent TypeScript module) and replacing the stub with your actual API call:

import httpx

async def get_weather(city: str) -> dict:
    async with httpx.AsyncClient() as client:
        resp = await client.get(
            "https://api.openweathermap.org/data/2.5/weather",
            params={"q": city, "appid": "YOUR_API_KEY", "units": "metric"}
        )
        data = resp.json()
        return {
            "city": data["name"],
            "temp_c": data["main"]["temp"],
            "description": data["weather"][0]["description"]
        }

Now test it. Add the server to your Claude Desktop or Claude Code config pointing at the local project, restart, and ask:

"What is the weather in Tokyo?"

If you get a real answer back, you're good. If not, check the MCP troubleshooting guide.

Step 4: Publish to PyPI or npm

Once your server works locally, tell your AI assistant:

"Build and publish this server"

mcp-creator runs the full pipeline:

1. Builds the distributable package (uv build for Python, npm run build for TypeScript)
2. Publishes to the registry (uv publish for PyPI, npm publish for npm)
3. Creates a GitHub repo and pushes your code

Your server is now installable by anyone:

pip install weather-mcp

That's it. From "I want a weather server" to pip install weather-mcp in a few minutes.

Need help with the publishing step? See How to publish your MCP server to PyPI for the detailed manual process and troubleshooting.

Step 5: Submit to MCP Marketplace

Every project mcp-creator scaffolds includes a LAUNCHGUIDE.md -- a checklist that walks you through submitting to MCP Marketplace. Follow it to get your server listed, security-scanned, and discoverable.

The short version:

1. Make sure your GitHub repo is public
2. Go to MCP Marketplace submission
3. Fill in the form with your package name, description, and repo URL
4. Submit for review

Once approved, your server shows up in search results and gets an install page with config snippets for every supported client.

Example: weather server, start to finish

Here's the full sequence. Real workflow, not a simplified demo.

Five minutes. One install command, a few sentences of English, and one function body worth of real code.

What mcp-creator generates

For reference, here is what the scaffolded project looks like:

weather-mcp/
├── pyproject.toml          # Package metadata, dependencies, entry point
├── README.md               # PyPI listing with install + config instructions
├── LAUNCHGUIDE.md          # MCP Marketplace submission checklist
├── .gitignore
├── src/
│   └── weather_mcp/
│       ├── __init__.py
│       ├── server.py       # MCP server with tools registered
│       ├── tools/
│       │   └── get_weather.py
│       └── services/
│           └── weather_service.py
└── tests/
    └── test_get_weather.py

Everything is wired together. The entry point is configured. The README has install instructions and config snippets. You fill in one file and publish.

Next steps

• How to build an MCP server -- the manual TypeScript SDK approach, if you want full control
• How to publish your MCP server to PyPI -- the detailed manual publishing process
• How to monetize your MCP server -- license keys, freemium, and remote hosting
• MCP Creator Python guide -- the full deep-dive on every feature
• MCP Creator TypeScript guide -- the full deep-dive for TypeScript
• Submit to MCP Marketplace -- get your server listed