Claude can't view images unless manually uploaded?

Cursor Version: 0.47.9
AI Model: Claude 3.7 Sonnet
OS: Windows 11

The Problem

I’m working with Claude in Cursor on a React Native project, and we frequently need to discuss UI issues visible in screenshots. I keep my screenshots in /misc/screens/ of my project directory.

If I upload a screenshot Claude can view it, no problem.

But manually uploading each screenshot to the chat whenever we discuss UI issues is tedious and interrupts my workflow.

I want to be able to tell Claude, “Check the latest screenshot”.

But when I make that request, or even when I explicitly reference screenshots by file path (e.g., \misc\screens\20250328_Expo_Go.jpg), Claude can’t actually view them, despite having agent access to my code files.

My Question

Is there any way to grant Claude the ability to view image files directly from my project directories?

The AI has full access to read my project files, but this seems limited to text files only. Are there any workarounds or settings I’m missing that would allow Claude to access and process image files without manual uploads?

I’d appreciate any solutions or alternatives that might streamline this workflow. Have any of you found a way to make this work?

I’m not sure about having it view screenshots in a folder, but you can drag and drop multiple images into the chat and send them. You can also click the image button and select multiples. Local file system only, though.

If you’re remote editing, the drag-and-drop won’t work. It needs to be from your local file system. Otherwise, it @mentions them, and I don’t believe it can see them when you do that. I’ve never tried.

I appreciate the quick reply.

Drag and drop is probably a bit easier, especially when sending multiple screenshots.

But imagine this scenario:

  1. You have your screen clipping utility save all screenshots to a project folder by default.

  2. You simply write, “Review the last 3 screenshots”

1 Like

Oh, don’t get wrong, it’s be awesome if they made that work.

What are the screenshots of, if you don’t mind asking? There’s multiple ways to reach the same end-result, there might be a better way.

Most of the time it’s a stylistic issue.

Me: Claude, fix the padding on element foo.

Claude: The padding is now fixed exactly as you requested.

Me: No, its worse.

Claude: Ah! I see the problem and now I’ve 100% fixed it for real.

Me: No, it’s back to exactly how it looked before I brought it up. Review this screenshot.

Claude: Now I see the issue. <actually fixes it>

Is it in a browser? then browser mcp would solve it by giving it access to the dom or dev tools to check

I would use this instead of screenshots:

Be sure to read 3c., you need to start the server before it will work.

Install locally, install the plugin manually, start the server, and add the MCP to Cursor; then, it can see all of the console logs, take screenshots, etc. You could also use fetch, although I’ve never had good luck with it. It constantly errors.

1 Like

Thanks, Mike.

this tool looks really helpful, and I will definitely keep it in mind for a future project.

But right now I’m building my first mobile app using React Native.

I also use Expo, so BrowserToolsMCP may have some limited usefulness with regard to Expo’s web viewer.

But my understanding is that the web viewer renders all of React’s code into HTML/CSS. There have been times that I have been able to figure out an issue by identifying some rendered CSS, for example, and reverse-engineering what it relates to in React’s code.

But, for the most part, when I take screenshots I am taking them on my Android device, and Windows Phone syncs the image to my laptop, and from there I have to manually save it and manually share it with Claude.

If such a tool existed for Expo Go it would really help me.

I am actually a web designer though, so it’s only a matter of time until I start building web projects in Cursor. And I will most certainly be thanking you for sharing this tool when that day comes around!

I know nothing about Android development, so this may not work. However, I was able to find this alternative. It appears to provide similar functionality as BrowserToolsMCP.

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.