Cursor needs awareness of the UI šŸ‘€

I wanted an MCP server without the need for a browser extension so I made this one recently which currently supports screenshots and getting the browser console log. I’m not suggesting it’s better than the other ones (and it is very new) but it does allow me to tell Cursor to verify the results in an actual browser (it uses Playwright) and also to e.g. tell it to fix the console errors on a page. It’s also possible to specify a specific CSS selector to focus the screenshot on.

100% agree here. We needs a built-in feature like this.

You could maybe have built this using AI already, sprint cycles of 2 weeks you could at least scaffold it, two modular UIs that see upstream and downstream dataflow with nodes that represent storage, auth etc. a bit like n8n. The other one can show the UI and you can select DOM elements just like that, or take snapshots and it knows what you are referring to.

Our Lord and saviour @leerob will surely make this happen, given his Next.js experience. :grin:

Yes, he’s good at it, let’s hope there will be a lot of useful things.

My early reaction here is: this would be helpful and useful, but also at the same time, the current way of working isn’t that bad!

Basically saying: lots of people have opinions about their editor/browser relationship. It’s hard to find the right solution for all of them.

I do think this is something we’ll explore more earnestly in time, but also wanting to set the expectation it’s not something we’re immediately working on.

W*****f just released browser preview. I’d like Cursor to have similar.

Hey, actually, Cursor already has a built-in browser, but it’s not very functional.

Thanks, I didn’t know this existed. But it seems to have no integration with Cursor agent.
Ultimately i’d like Cursor agent to be able to have access and communicate about UI without having to provide screenshots when doing frontend work

Considering how AI interactivity with images/video is developing now, this would be really cool. Especially along with a browser that’s built right into the Cursor, with all the data as context…
Sounds too sweet

This is a must, cursor!

I think it would be great if Cursor has a visual preview on what the product would look like, just like how lovable.dev has

like i wish that was a default thing so it would be easier for the user to see what is going on instead of jumping from A to B to C to D to E in order to see a preview of the product.

Hey, there is an integrated browser in Cursor. Have you tried it?

What? I didn’t know that!

That didn’t work. The feature needs to be a default. I need a feature that goes from point A to point B. No in between shenanigans of complications.the UI needs to make it easier to get to, it just needs to work.

Like come on. Adobe’s Dream weaver was already ahead of cursor on this! Lovable is ahead of cursor on this! Why is this too complex for cursor?

Yes, you’re right, the functionality is limited, but it’s not our development. It’s originally integrated into VS Code. Maybe in the future, we’ll create our own solution for this.

+1 for this!

Cursor is surrounded, yesterday Bolt has introduced it too:

I use Cursor as my main Ide it’s incredibly powerful and makes my development smooth. But there’s one key feature missing that really slows things down the ability to preview web apps and inspect DOM elements directly inside Cursor.

Right now, I constantly switch over to Windsurf just so I can inspect an element, and send it to the chat for changes. Cursor has everything else I need, but I have to leave it just to do this one thing. It breaks the flow, wastes time,

If Cursor could include a live web preview and let us click elements to inspect and send them straight into agent for editing, that would be a great. It would massively speed up debugging and design work, and really unlock more of Cursor’s potential.

I know a lot of us using Cursor would benefit from this. Let’s make it happen team cursor

The ability to preview your web app inside cursor, with the ability to inspect DOM components, and send that component code directly into Composer for the agent to iterate on.

Without this, we have to rely on sending screenshots, which is not always precise or clear with its understanding.