Hi everyone,
I’m exploring the ChatGPT Apps SDK and would love some guidance on best practices for building and deploying apps efficiently. Any tips, common pitfalls, or resources you recommend would be greatly appreciated!
1 Like
I’ve documented most of my pitfalls + lessons learnt here
But here is some more tips:
- A great way to get started will be trying out + running the example apps provided by the OpenAI team (GitHub - openai/openai-apps-sdk-examples: Example apps for the Apps SDK), this is how I started
- Understanding MCP and specifically tools + resources
- Bundling + serving your assets for your resources + widgets
- Adding these widgets + resources to your tool calls
Very high level tips, but OpenAI really provides nice docs + examples already but feel free to ask me any questions!
1 Like
A few things that reduced pain for me:
- Refresh the connector after any MCP change (tools/metadata), otherwise you’ll chase ghosts.
- Get tool annotations right early (readOnlyHint / openWorldHint / destructiveHint). Missing/incorrect hints seem to be a common rejection cause.
- Avoid iframes/frame_domains unless absolutely necessary (adds review friction).
- Test across web + mobile; handle errors cleanly; keep latency low.
- Minimize data collection/return (only what’s needed for the task + privacy policy).
Also: version widget URIs to avoid caching issues during iteration.
1 Like