Show HN: OpenAI Apps Handbook

6 months ago

Prerequisites include Python 3.10+, basic understanding of Python dataclasses, FastAPI/async Python, HTTP/REST concepts, and HTML.
An MCP server app has three core components: widgets, functions to respond to ChatGPT's requests, and FastAPI + Uvicorn serving.
Widgets require an HTML template, metadata, and a unique template URI.
Functions include list_tools(), list_resources(), call_tool_request(), and read_resource().
FastAPI serves GET /mcp for SSE stream and POST /mcp/messages for follow-up messages.
Step-by-step guide to create a project directory, virtual environment, and install dependencies.
Basic structure of main.py includes defining widget data structure and initializing FastMCP.
Adding widgets involves defining them in a list and building lookup dictionaries.
Input validation is handled using Pydantic models.
Tool metadata includes OpenAI-specific hints for rendering.
Handlers for list_tools, list_resources, and call_tool_request are registered.
Testing the server involves running it locally and using tools like MCP Inspector.
Deployment considerations include environment variables, Docker, and production setup with gunicorn.
Common patterns include authentication, rate limiting, database integration, and caching.
Troubleshooting tips cover server checks, widget validation, input schema, and CORS issues.

Hasty Briefsbeta