The IDE performance visibly degrades while agent generates content.
Steps to Reproduce
Ensure more than one conversation with some history open.
Prompt the agent to make code changes (agent mode /auto).
Try to perform another action in the IDE.
Examples:
A Ctrl+F for search in an open editor takes about 4 seconds to render the search box.
Even opening another conversation takes >4-5 seconds.
Occasionally, I even get the vs code prompt suggesting to Reopen the IDE due to unresponsiveness.
Expected Behavior
The IDE must stay responsive regardless of the agent actions.
I don’t have MCP tools enabled except one server for the Chrome dev tools.
The prompts that I use do not make use of it.
I’m not sure which is the exact set of conditions that cause this but it’s my daily experience now.
All I can share for certain is that it’s definitely related to the generation because as soon as it stops, the IDE responsiveness is back to normal. And that wasn’t like that before (weeks ago, can’t pinpoint a moment)
This doesn’t stop me from using cursor but I must confess that the experience with it severely degrades.
Hi @Georgi_Pavlov, thanks for the detailed report. A few things that would help narrow this down:
Process Explorer - While the agent is actively generating, open CMD + Shift + P > Developer: Open Process Explorer and check which process is consuming the most CPU (e.g., extensionHost, ptyHost, renderer, etc.). A screenshot of this during the lag would be very helpful.
Extension Monitor - Go to Settings > Application > Experimental > enable “Extension Monitor Enabled”, then CMD + Shift + P > Developer: Open Extension Monitor. This will show if any extension is contributing to the slowdown during generation.
Quick test - Can you try launching Cursor with cursor --disable-extensions and reproduce the issue? This will tell us if an extension is amplifying the problem.
Conversation history - You mentioned having multiple conversations with history open. Roughly how many conversations are we talking about, and how large are they?
This will help determine whether the bottleneck is in rendering, extensions, or the agent process itself.
Process Explorer shows most intensive activity in `cursor main > window’.
The extensions CPU activity is negligible - between 0-1% with very rare spikes to 2-4%.
2 and 3 i will contribute later if you still think it’s relevant based on the figures above
4 Conversation history - usually 3-5 open in parallel, but I tend to keep them pretty long. I observed hte context gets summarized multiple times within one conversation thread.
I had the same issue. Extremely slow rendering in the agent chat window. A task that would normally take 5-10 minute to generate code for took 60 minutes. Restarted cursor and that did not seem to help. Rebooted my macbook seems to have improved things (for now).
