I setup my MCP server and am serving the HTML files as resources.
I can see in the Playground that the resources are getting retrieved and that the tools are returning the appropriate URI:
_meta: {
‘openai/outputTemplate’: ‘ui://piccollage/carousel.html’,
However it is not displaying the UI. Is the Apps SDK UI functionality supposed to work already in the Playground, or only in the main chat interface?
Thanks for your help!
Or do I have to get special permission for this to work?
What you need to read is at the end instead of at the start of Connect from ChatGPT
Developer verification
Verification
All submissions must come from verified individuals or organizations. Once the submission process opens broadly, we will provide a straightforward way to confirm your identity and affiliation with any represented business. Repeated misrepresentation, hidden behavior, or attempts to game the system will result in removal from the program.
Ironically at the end:
Before you begin
Connecting your MCP server to ChatGPT requires developer mode access:
- Ask your OpenAI partner contact to add you to the connectors developer experiment.
- If you are on ChatGPT Enterprise, have your workspace admin enable connector creation for your account.
- Toggle Settings → Connectors → Advanced → Developer mode in the ChatGPT client.
Once developer mode is active you will see a Create button under Settings → Connectors.
Those steps:
A click-through to ‘Advanced Settings’.
Create.
Try.
Did you get that far?
Then go back and read everything expected, everything about being a partner-level brand-name product that would ever be reviewed and featured before you work too hard at a process not even opened yet.
+1 @Jaime_Cham , I only get structured outputs in developer mode
Yes, I got that far. I understand that they are only letting the big guys in to have it be in production, but for the sake of developer mode, since they opened up the API and want people to start developing, it’s vague enough that maybe it does work for developers, so I wanted to see if anybody else has got it to display the UI yet. Thanks for the feedback!
Yes it is possible in developer mode to render the UI. I’ve got something to work (see screenshot below)
However i am yet to find an easy way to iterate and test in dev environment.
I’ve used the inspector as per the guide but they are mainly to render the tool raw response and not the UI
The ChatGPT client also caches the UI component, where there is a need to refresh everytime a UI change is made.
If anyone knows of a way to test and iterate UI components locally while connected to data from ChatGPT, would love to learn how to do that.
I’m also having an issue with displaying UI on a working MCP server.
Both through the playground and also using developer mode in chatGPT.
In both cases, I get a valid response from the tool call, but the HTML never renders inline. Tried everything. Any tips?
I’m following up here with more details because what I’m seeing from other developers suggests this may be an issue with the ChatGPT platform.
My room responses is indicating success but the UI widget is just not loading. Here’s the full response…
{“_meta”:{“openai/outputTemplate”:“ui://widget/video-generator-v10.html”},“content”:[{“type”:“text”,“text”:“Hello world widget opened”,“annotations”:null,“_meta”:null},{“type”:“resource”,“resource”:{“uri”:“ui://widget/video-generator-v10.html”,“mimeType”:“text/html+skybridge”,“_meta”:null,“text”:“Hello world”},“annotations”:null,“_meta”:null}],“structuredContent”:{“status”:“launched”},“isError”:false}
Would love to be wrong - anyone else dealing with similar?
I can’t render anything dynamic for the life of me. I can render simple divs and whatever, but whenver I want it to depend on an actual api, it just doesn’t render anything. no error, nothing.
Even I tried to create an app in developer mode. Here is what I did
Now, here I have connected to my MCP. My MCP is even responding with UI but still the ChatGpt on my mobile App is showing the normal UI (not the widget).
Does anyone know anything about this app.json? Read somewhere that we need to configure the app.json URL.
Or, instead if new connection is there something like `New App`?
I’ve been playing with it and made a simple UI that reads input to an MCP tool via window.opeai.toolInput. I’ve written a tutorial on it - Building UIs inside ChatGPT with OpenAI Apps SDK and Node.js: Part I
Thank you for sharing @mikeborozdin , what was the quickest way you found to refresh the template? There are times i found that clicking “refresh” here works, but sometimes it doesnt, and i end up having to delete and reinstall the app
Also, to test template on chatgpt is too cumbersome, i am still unable to figure out a way to test these templates locally. Let me know if you do have a method
hey @Jaime_Cham, the UI features won’t render in the Playground, they only load in ChatGPT when connected to a live MCP endpoint.
If you just want to get one running quickly, there’s a short guide here on deploying an MCP server to the cloud: Deploy and host a ChatGPT App - mcp-agent
That’ll give you a proper public /sse endpoint that connects cleanly to ChatGPT Developer Mode.
I had the same issue. UI features do not work on Business/Enterprise licenses, even in Dev mode. I switched to a Pro account and the connector immediately worked with the correct visual inside the ChatGPT UI.
ChatGPT Apps are not yet supported with Business/Enterprise licenses, see Connect From ChatGPT:
Note that ChatGPT Apps is only supported on Plus, Pro, Go and Free plans. Business, Enterprise, or Education plans do not support ChatGPT Apps today, though availability for those plans is on our roadmap. You will still be able to use developer mode on those plans, but you will not be able to integrate Apps into ChatGPT.
