Cupcake MCP Server + Wasmer

This example shows how to run a Model Context Protocol (MCP) server for ChatGPT on Wasmer Edge.

ℹ️ MCP servers connected to ChatGPT should expose at least two tools—search and fetch—so ChatGPT can both discover content and then retrieve specific items.

Demo

https://mcp-chatgpt-starter.wasmer.app/sse

Add it to ChatGPT as a connector (no auth), and then just ask ChatGPT to interact with it:

How many cupcakes Alice ordered?

How it Works

All logic lives in server.py, but you can think of it in sections:

Data Section

The server loads cupcake records from a local records.json file and builds a lookup dictionary:

RECORDS = json.loads(Path(__file__).with_name("records.json").read_text())
LOOKUP = {r["id"]: r for r in RECORDS}

Models Section

We define Pydantic models to structure responses:

SearchResult and SearchResultPage for search results.
FetchResult for full cupcake order details.

Tools Section

Two MCP tools are exposed via FastMCP:

search(query: str) Splits the query into tokens, performs keyword matching across title, text, and metadata, and returns a list of matching results.
fetch(id: str) Retrieves a single cupcake order by ID from the lookup dictionary and returns full details, including optional url and metadata.

Entrypoint Section

At the bottom of server.py, the app is created and run:

app = create_server()

if __name__ == "__main__":
    app.run(transport="sse")

The server uses Server-Sent Events (SSE) to communicate with ChatGPT’s MCP integration.

Running Locally

Install dependencies:

pip install -r requirements.txt

Run the server:

python server.py

Your MCP server will now be running and ready for connections from an MCP client (like ChatGPT with MCP enabled).

Example Tools in Action

Search tool (search("red velvet")) Returns a list of cupcake orders that mention “red velvet.”
Fetch tool (fetch("42")) Returns the full details of order 42, including text, metadata, and an optional URL.

Deploying to Wasmer Edge (Overview)

Include both server.py and records.json in your project.
Deploy to Wasmer Edge, ensuring the entrypoint is server.py.
Access it at: https://<your-subdomain>.wasmer.app/sse

How it Works

All logic lives in server.py, but you can think of it in sections:

Data Section

The server loads cupcake records from a local records.json file and builds a lookup dictionary:

RECORDS = json.loads(Path(__file__).with_name("records.json").read_text()) LOOKUP = {r["id"]: r for r in RECORDS}

Models Section

We define Pydantic models to structure responses:

SearchResult and SearchResultPage for search results.

FetchResult for full cupcake order details.

Tools Section

Two MCP tools are exposed via FastMCP:

search(query: str) Splits the query into tokens, performs keyword matching across title, text, and metadata, and returns a list of matching results.

fetch(id: str) Retrieves a single cupcake order by ID from the lookup dictionary and returns full details, including optional url and metadata.

Entrypoint Section

At the bottom of server.py, the app is created and run:

app = create_server() if __name__ == "__main__": app.run(transport="sse")

The server uses Server-Sent Events (SSE) to communicate with ChatGPT’s MCP integration.