when I take screenshot using cmd + shift + 4 and drag the image it does not attached to agent chat in cursor 3 but when I save to clipboard it then attach to chat easily for more context I have attached the video.
Steps to Reproduce
take a screenshot in cursor 3 in full screen in web browser mode in cusror then try to drag that image into chat it will not able to drag but when you save to clipboard and then cmd +v it attaches to chat window.
Hey, thanks for the report and the video. This is a known limitation of the new Agents Window Glass layout. Drag and drop for files and images into chat doesn’t work reliably everywhere yet, unlike the regular editor layout. I can’t share an ETA for a fix right now.
The good news is you already found a workaround. Pasting from the clipboard with Cmd+V works reliably for screenshots, so I’d recommend using that for now.
We’ve logged the issue on our side. If there’s an update, I’ll reply in the thread.
Clipboard paste is the one I landed on too, the drag layer has been hit or miss for a while. The thing that still costs me time is after the image is in: the agent picking which element I meant on a dense screen. Do you get clean targeting from the screenshot alone, or end up spelling it out in text?
Hey, thanks for sharing. Clipboard paste is still the most reliable option for now since drag and drop in Glass still has limitations. We’re tracking an issue on our side, but there’s no ETA for a fix yet.
On targeting: it’s hard to expect perfect accuracy from a single screenshot on a busy screen. The model sees the whole image, and without a hint it can pick the wrong thing. What helps:
You can draw a box around the area you want directly in Cursor on the attached image, then the agent knows exactly which element you mean. One important note: this works in the regular editor window, but not in the Agents Window (Glass) yet.
If area selection isn’t available, add a short text hint with the image, like “the Save button in the top right” or “the second card in the list”. Just a couple of words can noticeably improve accuracy.
For web and UI, try Design Mode. You can select an element directly in the preview, no guessing from the screenshot. More details: Design Mode | Cursor Docs
Let me know if there are any details in your case and I can help more precisely.
Thanks Dean, clearest summary I’ve seen of what actually helps. The thread running through all three is the same: give it the element, not just the picture. The gap I keep hitting is that your two precise options are surface-bound, draw-a-box is editor-only (not Glass yet), and Design mode is web-only. So native apps and the Agents window fall back to the text-hint guess. That’s the case I’ve been solving by carrying the marked element plus intent as structured text the agent reads, whatever the surface. Any plan to bring box-on-image into the Agents window? That would close most of it.
You nailed it. Right now, both precise options are tied to the surface: draw-a-box works in editor-layout, and Design Mode is only for web. In the Agents Window (Glass), box-on-image isn’t available yet, and that’s a known parity gap between Glass and the regular editor.
I can’t share a specific plan or ETA for bringing image annotation to the Agents Window right now. Glass is gradually catching up to editor-layout feature-wise, but it wouldn’t be fair to promise a timeline or even that this exact capability will ship.
Your approach of sending the annotated element plus the intent as structured text that the agent can read regardless of surface is a solid workaround today, especially for native apps where neither draw-a-box nor Design Mode applies. For web, I still recommend Design Mode since it lets you select the element precisely without guessing from the image: Design Mode | Cursor Docs
If there’s an update on box-on-image in Glass, I’ll reply in the thread.
Appreciate the straight answer, Dean, and for naming the native gap instead of hand-waving it. I’ll keep sending the annotated element plus intent as structured text for native, and Design Mode for web. If box-on-image lands in Glass, that closes most of it.
We’ve made some improvements to drag and drop in the agent chat since you posted. Could you update to the latest version and give it another try? Quick check:
Take a screenshot with Cmd + Shift + 4
Drag the image into the agent chat input
It should attach to your message
Let me know if the screenshot still won’t attach when dragged, any updated details would help. Pasting with Cmd + V remains a reliable fallback either way.
One thing worth ruling out in the repro: dragging the floating screenshot thumbnail from the corner of the screen before it saves is a different drag path than dragging the saved file out of Finder. The thumbnail promises the file rather than handing over a path, and plenty of apps only accept the second. If that is what Sehaj was doing, the fix may not cover it.