MCP results are not rendering in the product list widget

Our company’s app “Cafe24” has recently been approved and officially published.
However, we are experiencing an issue in the published environment where the widget does not receive or render MCP (tool) results.

  • The app status is Published.
  • The MCP tool executes successfully and returns valid results.
  • Despite this, the widget UI remains empty and does not render the MCP output.
  • In addition, the widget continues to be loaded from the following sandbox domain, which remains in a pending state:
    https://…sandbox

We would appreciate clarification on the following points:

  • Why the app continues to use the web-sandbox.oaiusercontent.com domain despite being in a Published state.
  • Why MCP (tool) results are not being rendered in the widget.
    This issue did not occur during development or pre-publication testing.

I’m running into a similar issue as described in this thread and wanted to share my observations to see if others have seen the same behavior.

I have a version of my app where the widgets were rendering correctly. Without any changes on the app, the widgets suddenly stopped rendering altogether - no UI, no visible error, just missing widget rendering.

At first I suspected a code issue, but here’s the interesting part - after deleting the app, clearing my entire cache/state, and creating a fresh connection to the app, the widgets started rendering correctly again.

I am still wondering whether it is something like a stale widget metadata issue, or issue with previously open session state. I’m not sure if previously open sessions, or SDK-side caching could cause the widget to silently fail. I wasn’t able to find any clear documentation explaining why this happens, so I’m curious. Any insight from others (or the OpenAI team) would be really helpful. Thanks!

1 Like

It sounds to me like widget/app code is getting stale, but it’s hard to tell just from these descriptions. If anyone can provide a minimal reproduction, it would be helpful for the team. I’ll file a ticket to take a look.

Thanks @benlesh for raising and @shkim30 for writing this up. We’re looking into it.

1 Like

@shkim30 do you see errors in your console when running your app?

Hello, @coreyching I have submitted the same issue (- Case Number: 05037410) to support@openai.com.

It is reproduced by all users.
It seems that the OpenAI Apps widget format is different from the resources currently provided by MCP.
Would it be possible to scan using the current MCP resources on the version 1.0.0 app, or alternatively, could you approve the already submitted version 2.0.0?

(It is inferred that version 1.0.0 probably scanned the widget resource which had issues.)

[Current Status]

MCP output is not visible in this widget.

Hi @coreyching ,
just checking in to see if you had a chance to review the issue below.

We’re currently blocked since the MCP output is not visible in the widget, so any guidance or next steps would be really appreciated.

Thanks in advance!

1 Like