About a month ago I was experimenting with giving Claude Code access to a browser. I was surprised by how good it was. AI can now both write apps and use a web browser to see if everything works. I thought this was huge. We can now digest data, write specs, implement features AND get feedback on them in a matter of minutes with AI. And I couldn't find much work being done on the last part. Why not build a product around AI UX testing?
A week later, ClankerView was born.
You enter a URL to your product and spin up 1-6 agents to review it, then watch as they get personas based on your target audience, open up a browser, click around, scroll, fill forms, and react to screenshots of what they see. They then write structured UX reviews that you can use to make improvements to your product.
You can even review behind authentication. Give the agent credentials in the instructions or skip the auth by pasting session cookies to the agent (they are deleted after the review).
I naturally tried it on all of my side projects and found plenty of useful insights, similar to what people who tried them told me.
I then tried a bunch of other products. For example, Framer's onboarding review found some useful actionable issues: the onboarding survey answers are ignored, it bothers you with the desktop app before you get to the web app, you land in a blank canvas after choosing a template, and a bunch more. The verdict was "Framer's marketing says it's for everyone. The product says it's for designers." 5/10 https://clankerview.com/share/UVCSqrBipcsxfxJh
Interesting — how does the agent authenticate to paid/protected endpoints? This is the piece most agent UX tools skip. HTTP 402 + macaroon tokens handle this natively without API keys.
Currently you can simply add a username/password to log in with under "+ Instructions" or log in yourself and paste in a JSON exported from the Cookie-Editor browser extension under "+ Cookies".
I'll look into HTTP 402 + macaroon tokens, thanks!
Hey HN! I'm Andri, the guy behind ClankerView.
About a month ago I was experimenting with giving Claude Code access to a browser. I was surprised by how good it was. AI can now both write apps and use a web browser to see if everything works. I thought this was huge. We can now digest data, write specs, implement features AND get feedback on them in a matter of minutes with AI. And I couldn't find much work being done on the last part. Why not build a product around AI UX testing?
A week later, ClankerView was born.
You enter a URL to your product and spin up 1-6 agents to review it, then watch as they get personas based on your target audience, open up a browser, click around, scroll, fill forms, and react to screenshots of what they see. They then write structured UX reviews that you can use to make improvements to your product.
You can even review behind authentication. Give the agent credentials in the instructions or skip the auth by pasting session cookies to the agent (they are deleted after the review).
I naturally tried it on all of my side projects and found plenty of useful insights, similar to what people who tried them told me.
I then tried a bunch of other products. For example, Framer's onboarding review found some useful actionable issues: the onboarding survey answers are ignored, it bothers you with the desktop app before you get to the web app, you land in a blank canvas after choosing a template, and a bunch more. The verdict was "Framer's marketing says it's for everyone. The product says it's for designers." 5/10 https://clankerview.com/share/UVCSqrBipcsxfxJh
Here's an agent reviewing HN because why not: https://clankerview.com/share/sv20WB2Ssd8h8Q9a
I genuinely think ClankerView can help everyone working on any sort of product that is accessible via a web browser.
I'm happy to answer any questions. Please let me know if you would use this and if not, why not?
That HN agent review is pretty usable. Very cool!
Thank you!
Interesting — how does the agent authenticate to paid/protected endpoints? This is the piece most agent UX tools skip. HTTP 402 + macaroon tokens handle this natively without API keys.
Currently you can simply add a username/password to log in with under "+ Instructions" or log in yourself and paste in a JSON exported from the Cookie-Editor browser extension under "+ Cookies".
I'll look into HTTP 402 + macaroon tokens, thanks!
[flagged]