Hi,
I would like to report what appears to be a regression with image/screenshot attachments in Cursor Chat.
For the last few days I have been able to take a screenshot and paste it directly into the Cursor chat using Ctrl+V. This has worked normally and has been very useful in my work, particularly when discussing visual/UI problems with the agent.
Today, however, this suddenly stopped working. Screenshots pasted into the chat no longer reach the agent.
Nothing has changed in the way I am taking or pasting the screenshots. The same screenshots can still be pasted and sent normally in ChatGPT.
Since direct screenshot copy/paste worked correctly in Cursor for several days and has now stopped, I suspect this may be a regression rather than an unsupported feature.
I know I can save screenshots as files and place them in the project folder as a workaround, but that adds unnecessary steps. Direct Ctrl+V into the chat is considerably faster and more useful during visual debugging.
Could you please investigate whether something has changed with image clipboard/attachment handling in Cursor Chat?
Thank you.
Steps to Reproduce
by using clip/paste in windows, Corel Capture
Operating System
Windows 10/11
Version Information
Cursor IDE 1.x.x (Windows 10)
For AI issues: which model did you use?
Not primarily an AI model issue – Chat UI paste/attachments
Hey, thanks for the detailed report. This looks like something local on the client side, not a general outage, so let’s narrow down the cause. A couple of quick checks:
After Ctrl+V, do you see an image preview chip in the input field? Or does nothing get pasted at all?
Does Reload Window Ctrl+Shift+P > Developer: Reload Window help, or a full restart of Cursor?
You’re using Corel Capture. Try taking a screenshot with the standard Snipping Tool Win+Shift+S and paste it. Does that work? This will help confirm if the clipboard source is the issue.
Does drag and drop of a PNG file straight into the chat input work?
Please share your version from Help > About and where exactly you’re pasting, in the Agents window or the built-in IDE chat.
Based on that, we’ll know where to dig next. Let me know how it goes.
I tried the troubleshooting steps you suggested, including restarting/reloading Cursor and using other ways of getting the screenshot into the chat, but none of them solved the problem at the time.
The interesting part is that today image copy/paste suddenly seems to be working normally again. I have not knowingly changed anything that would explain this.
When the problem occurred, the same screenshots and the same clipboard workflow continued to work normally in ChatGPT, while they would not paste into Cursor Chat. Copy/paste had also been working normally in Cursor for several days before it suddenly stopped.
So the sequence has been:
Image copy/paste in Cursor worked normally for several days.
It suddenly stopped working.
The same screenshots could still be pasted into ChatGPT.
I tried the suggested troubleshooting without success.
Today, without any deliberate change on my side, image copy/paste in Cursor appears to be working normally again.
I am using Corel Capture most of the time. If the problem returns, I will check specifically whether the preview chip appears after Ctrl+V and compare Corel Capture with Win+Shift+S.
At the moment I cannot explain why it stopped or why it started working again, but I thought this information might be useful in case this is an intermittent issue.
Thanks for coming back with an update. The fact that paste started working again on its own, without any changes on your side, points to something local and intermittent rather than a widespread issue. Good news that it’s working now.
If the issue comes back, it’d really help to capture a couple things right at the moment it happens:
After Ctrl+V, does the preview chip show up in the input box right away, or does nothing get pasted at all.
Try the same screenshot from Corel Capture and from the built-in Snipping Tool Win+Shift+S. Does one paste differently than the other. This helps tell if it’s related to the clipboard source.
Does dragging and dropping a PNG file directly into the chat input work.
The version from Help > About at the time it happens.
With those details we can dig deeper. Let me know if it repeats.
attachment pipeline (broken). You paste/drag into chat → Cursor must copy the file into assets/ → I read the copy. The copy step never executes. Source location irrelevant — Desktop, wherever. Pipeline dies at the write.
Hey, thanks for the detailed report. I can see in the screenshots that the agent can see the source file Screenshot.png on Desktop (10:18), but the copy Screenshot-<uuid>.png never gets written into the assets/ folder. That means the attachment write step isn’t running. This looks different from what was described at the start of this thread (that one was intermittent clipboard paste), so I want to dig into this separately.
To go deeper, can you share:
Your version from Help > About and your OS.
Is the project local, or is it remote / SSH / cloud agent / private worker? For those surfaces, the path to assets/ behaves differently.
The Request ID from one of the sessions where the attachment didn’t get written (chat context menu top right > Copy Request ID). If Privacy Mode is enabled, keep in mind the ID might not be available.
Do you see the preview chip in the input right after Cmd+V or drag, or does it never show up?
As a temporary workaround, the agent can read files directly from disk, so you can save the screenshot into your project folder and reference it via @file. It’s not as convenient, but it works while we investigate.
Let me know, and we’ll see exactly where it breaks.
Hey, thanks for coming back with details. The version and the fact that the chip shows up already help a lot.
Quick clarification on how it works: when you paste or drag an image into chat, it gets sent to the model directly as visual context. The agent does not need to find a file in an assets/ folder to see it. So if the agent goes looking for Screenshot-<uuid>.png in the project and says “attachment failed to write to disk”, that usually means the image did not reach the model as an attachment, not that writing the file failed. The project assets/ folder is not the right place to look for it.
To figure out where the image is getting lost, I need the Request ID from the session where you pasted the image, the chip appeared, but the model did not see it:
Request ID: chat menu in the top right corner (three dots) > Copy Request ID. If Privacy Mode is on, the ID might not be available. If so, let me know whether it is enabled.
Which model are you using in that session, or are you on Auto? Some models and routing handle images differently.
Does Reload Window Cmd+Shift+P > Developer: Reload Window or a full restart help, even temporarily?
As a temporary workaround: the agent can read files directly from disk, so you can save the screenshot into your project folder and reference it via @file. It’s less convenient, but it works while we dig in.
With the Request ID, I can check whether the image bytes reached the backend and investigate further. Let me know.
Hey @adrienlo, thanks for the Request ID. That let us trace everything end to end.
First the good news. Image upload is working fine on your side. The preview chip is real, and the image did reach our servers in that request, so nothing is broken on your end.
The issue is the model. This session was using Kimi K3 Max, and it currently runs as text-only. For text-only models, Cursor usually sends the image to a separate vision model to generate a text description. We found that for Kimi K3 this step isn’t running right now. As a result, the model gets no visual context and starts guessing, including making up a story about a file in assets/ that doesn’t exist. I’ve shared this with the team.
For now, images should work normally if you switch the session to a vision-capable model, like Claude, GPT, Gemini, or Grok. If you try the same screenshot on one of those and it describes it, that also confirms it’s a Kimi K3 routing issue. Let me know how it goes.
