I’m experiencing the same issue.
When I access our ChatGPT app using a free ChatGPT account, after opening it, I notice that the conversation does not seem to enter or route into our ChatGPT app, even though I have already selected the app.
However, if the user is on Plus or another paid plan, the conversation enters the app as expected.
For free-tier users, it seems that only when I explicitly say something like “call the xxxx tool” in the conversation does it actually enter the app/tool flow. Otherwise, the conversation does not go into the ChatGPT app.
Is this behavior expected? Has anyone else encountered this issue?
+1 happening with us! Anyone have any info on this bug? How do we notify the team?
Same here! It’s not calling the MCP until you explicitly tell ChatGPT to do it
Same here, problem observed on almost all of the apps in the store where web search has a minimum relevance (doesn’t happen for very private data apps, like spotify for instance, because websearch has no business taking over the use of the app).
i wondered it tooo
i can’t us my app in GPT free
Hi all,
Thanks for flagging this. We've raised this with our team and will update here as soon as we receive an update from them.
~Smith
It seems that today support for ChatGPT apps on free plans is gone.
This is fine, but the UI still suggests that app can be installed but the model clearly does not see them. [screenshot]
This is pretty bad for developers as people might think the app does not work while it does.
Any fix expected ?
EDIT: someone reported that some apps do work while other do not but could not verify any app working.
Hi all, we’re aware of this issue and are looking into it. If you’re comfortable with it, sharing conversations where this occurred would be helpful for us to pinpoint the issue.
Hi @casey-chow, thanks for looking into this. Here are two conversations from a free-tier account using our PagesJaunes app.
Conversation 1 (conv1.png) — App never invoked, even after being called out: The user asked to find a plumber near Cesson-Sévigné. ChatGPT answered using generic web search results, ignoring the PagesJaunes MCP app. When the user explicitly asked “why didn’t you use the PagesJaunes MCP app server I mentioned?”, ChatGPT acknowledged the mistake and apologized — but then fabricated a response pretending to have used the PagesJaunes app, while actually still relying on its own web search. The app was never truly invoked at any point in the conversation.
Conversation 2 (conv2.png) — App invoked, but only on the second attempt: Same type of request. The app was not triggered on the first message. Only after the user explicitly challenged ChatGPT did the app actually get called and return real PagesJaunes results.
Pattern observed:
On free-tier accounts, the app is consistently skipped on the first invocation, even when pinned in the input bar
The model silently falls back to web search
When called out, it may either hallucinate app usage (conv1) or finally trigger correctly (conv2)
On paid accounts, the app triggers correctly from the first message
The hallucination case in conv1 is particularly concerning from a developer perspective, as end users have no way to detect that the app was never actually called.
Hope these help. Happy to provide more test cases if needed.
–
One more thing I forgot to mention:
We also tested with web search disabled in the user settings — in that case, the PagesJaunes app is invoked systematically and correctly from the first message.
This strongly suggests the regression is not about app invocation being broken per se, but rather a prioritization conflict between web search and MCP apps: when both are available, web search consistently wins, even when a specific app has been explicitly selected or mentioned by the user.
Hope this additional datapoint helps narrow it down.
Hey everyone, Thanks for flagging our team is actively looking into this issue. We will update very shortly. Thanks!
Thanks for sharing examples and screenshots as well they help us immensely!
I can confirm on my end that disabling web search from user settings solves the problem!
I think this is relative to model changes within the conversation.
I start with the app installed, I ask a few questions and the model correctly recognizes the App. After a few messages (crucially after usage limit banners appear) the model does not invoke the app anymore and if I ask the model which apps it has access to it says None (as screenshot in previous post).
It looks like the model gets swapped but the MCP context is not injected in the new conversation with the new (worse) model (which would be in principle able to tool call but does not have the app loaded in context).
I’m the OP of the thread, in my experience it almost always fails already from the first turn, even when you explictly tag the app, see screenshots I posted. Disabling web search consistently solves the problem.
Hey all, we deployed a change that we believe should help a bit with this and are trying to assess whether it’s working. If you still can’t invoke apps on the free tier please let me know.
Same issue still - tested across 3 different apps with a Free tier account and behaviour was the same.
