MCP tool calls only fire when the app is explicitly selected

Wanted to check if anyone else is seeing this.

Expected behavior: When a user mentions the name of the app in their prompt, ChatGPT should recognize the app, pick the right tool, and invoke it automatically, without the user needing to select the app from the picker first. This used to work reliably for us.

What we’re seeing now: ChatGPT does recognize the app, the chat trace shows “Looked for available tools” and “Thought for Ns”, but no tool actually fires. Instead, the model responds in plain text, asking the user for the inputs it would have passed to the tool.

What still works: If the user explicitly selects the app from the picker and resubmits the prompt, every tool fires normally.

Wanted to check if anything changed recently, or if this is a known issue that the OpenAI/ChatGPT team is working actively to mitigate?

Interesting thread ,I hit your post while looking into how invocation behaves for unselected apps, so I tried to reproduce this today (Aug 30) and couldnt.

Setup: a dev-mode connector (personal notes app, 3 read-only tools, no auth), ChatGPT Plus, GPT-5.6 Sol, memory off, custom instructions blank, fresh chat per trial, first submission counts.

  • installed but NOT selected in the picker, prompt mentions the app by name: 12/12 tool calls fired
  • explicitly selected from the picker: 10/10
  • same as the first arm but with a rewritten tool description: 12/12

0 cases of “Looked for available tools” with no call, and 0 cases where it asked me for the inputs instead of calling the tool. every fired call is confirmed in server logs, not just the UI.

So at least for a dev mode connector on Plus this looks fixed since June. Caveats: only one app was installed (nothing competing for the name), the prompts named the app explicitly, and its one account/model/day — directional, not a guarantee it behaves the same for your app or a published one.

Happy to share the raw trial CSV if thats useful. Did this resolve on your side too, or are you still seeing it?

Thanks for taking the time to reply. This used to be a problem for previous/older versions of ChatGPT. This problem seems to be resolved for the newer versions.