THIS GUY ADDED A LIVE BROWSER TO CLAUDE CODE SO YOU CAN CLICK ANY ELEMENT AND EDIT IT INSTANTLY biggest issue with vibe coding UI is that you have to describe what you want to change. if you prompt it the wrong selector or wrong component, Claude can't find... it. now you just click it. your app runs in an embedded browser with Claude Code. you can click any button, any text, any div. Claude instantly knows exactly what you're pointing at. click. instruct. done. no more "change the button in the top right corner of the second card component." all you have to do now is just click the button. AND its open sourceshow more

Hey, I'm the creator. Really appreciate the shoutout 🙏 timing is wild - shipped a new version yesterday that adds flow recording, so you can capture a full click-through of your app for many purposes. Feedback and contributions are welcome

@grok analyze all information in the video above and find the official github repo

the biggest time sink in claude code isn't the model thinking, it's me trying to explain which button i mean using prose. pointer beats paragraph every time

can do something like this but better. or you develop a proper AEP (agent element protocol) but you don't have intelligence for that.

Same thing, right?

pointing at the thing you want to change is how humans have communicated since caves. describing it in text to an AI was always a weird detour

Isn't there a better version of this with @benjitaylor's Agentation? I feel like Agentation is more cleaner, open source and still gives way more context for comments to agents.

That’s really smart. I’m juggling the idea of making something that extends Claude usage limits by working with the caching and token usage. Most people don’t know that if you don’t use Claude within five minutes of your last prompt, the next prompt it completely reloads its memory and all the code it’s looking at. So thousands of tokens are going to waste. Should I build it?

built lock in with claude - describing layouts killed 40% of my time. curious if the click accuracy beats text prompts tho

This fixes the biggest pain in vibe coding. Describe wrong = stuck.

we are mass-producing senior frontend engineers in a weekend and agencies are still charging $150/hr for someone to argue about padding values in figma the gap between what one person can build alone and what a 10 person team ships is closing so fast it's actually terrifying for anyone whose entire business model is "we have more people"

This is something that I'd love to have natively in ClaudeCode. @AnthropicAI

can't you do this too in the claude code preview section?

@om_patel5 this is a game changer. finally more freedom to tinker and adjust on the fly without typing a single line

Repo:?

Did someone test it?

been waiting for someone to do this. the gap between "I see the bug" and "I can describe the bug to the LLM" is where half my dev time goes. click-to-target kills it. open source is the cherry on top

Glass has this built in for Claude as well as Gemini and GPT

u won't believe how long i’ve been looking for this

whoa that's genius!! no more selector hell

Does this work with dynamically generated components too? I'm thinking about cases where the DOM elements change based on user state or API responses - curious if the selector mapping stays accurate.

does inspect element not already exist 😭

It’s that a native feature now

wow this is huge @ProEvilz codex needs this

Embedding your app in a browser and pointing at elements makes Claude Code instantly aware of context

Yeah this is huge. The selector description problem is brutal - you're trying to explain which div in a nested tree and Claude's just guessing. Half the time it clicks the wrong thing or can't find it at all. Having a live browser where you can actually point at stuff is so much better than trying to describe a selector in natural language. Makes me wish more agent tools had something like this baked in.

The gap between "I want to change this" and actually changing it just went to zero. We went through the same thing building internal tools at AutoDS. Engineers spent 40% of their time just explaining what needed to change to other engineers. The moment you remove the translation layer, speed doubles. Every time.

Sure, but where’s the repo?

Is it better than Agentation? That one has been a massive time saver for me.

Only for Claude? Pity there isn’t something like this for Codex

💀