Feature request for product/service
Cursor CLI
Describe the request
Hi team,
I’d love to see support for MCP servers sending images directly into a chat, without the user having to manually hit “Send.”
This would allow tools to generate or capture images — for example, a screenshot of a UI, a visual diff, or a rendered layout — and post it straight into the conversation.
The key idea is that the MCP should be able to push an image message on its own, just like a user-attached image (thumbnail, persistent, quotable, and viewable).
That way, an agent could run a task, screenshot the result, and immediately “see” the outcome — enabling true visual feedback loops.
Example workflow:
The MCP performs a change (e.g., edits code or updates a layout).
It takes a screenshot or generates a new image showing the effect.
The image is sent directly to the chat by the MCP.
The agent can then interpret that image and continue reasoning visually.
Safety and handling could stay fully controlled:
Only image types allowed (PNG, JPEG, WebP).
Size limits enforced (e.g., 10 MB per image).
Images passed as local files or base64 text.
Cursor decodes and copies them into its sandbox before display.
Why this matters:
It enables real multimodal feedback between tools and the AI —
the agent could make a change, request a screenshot through the MCP, and immediately analyze the visual result — without the user needing to manually intervene or re-upload images.
Thanks,
Paul (RayLx)