RiqochLovable, Cursor, Claude Code: how to get feedback on an app built with AI
Your app was built in a few prompts. How to have your team or a client try it, and collect precise feedback, whichever tool built it.
With Lovable, Cursor or Claude Code, a working app can come out in a day. What hasn’t sped up at the same rate is the moment other people look at it: a co-founder, a designer, a client, the first users. And that is often where it all happens: an app generated quickly needs human eyes to find out what is off.
Why it is harder than it looks
The app isn’t always online. With Cursor or Claude Code, it often runs on your computer. To show it, you have to deploy it somewhere, or share your screen on a call.
Each tool has its own comments, or none. Lovable lets people comment on its apps’ preview, and those comments go to its AI. But if your project moved from Lovable to Cursor, or part of it was written by hand, you need one shared place.
Feedback is vague. “The home page feels a bit empty”: useful, but where exactly? To fix things quickly, with or without AI, you need to know which element it is about.
Versions keep coming. When the AI produces a new version in ten minutes, the screenshots sent yesterday are already out of date.
What a good feedback loop needs
- A link to the real app, that anyone opens in their browser, without an account.
- Comments placed on the app’s elements, not on a picture.
- A link that follows the versions, so reviewers don’t go looking for the latest one.
- A short path from feedback to fix, ideally all the way to the coding agent that will make the change.
With Riqoch, whichever tool you use
Riqoch doesn’t build your app: it gets it reviewed, whichever tool built it. Lovable, Cursor, Claude Code or by hand, it is the same code in the end.
If your app comes from Lovable, sync the project with a GitHub repository from Lovable. In Riqoch, connect GitHub once, pick the repository and the branch: Riqoch fetches the project, installs the dependencies and keeps it up to date.
If you work with Cursor or Claude Code, just open the project’s folder on your computer, or its GitHub repository. Riqoch recognises Vite projects (React, Vue, Svelte…), Next.js, Nuxt, Astro, SvelteKit, Remix, Angular and most JavaScript projects, monorepos included, as well as Django, Flask, Streamlit, Rails and PHP.
From there, it is the same for everyone:
- Riqoch runs your app, with nothing to deploy.
- You set which screens to look at, and in which order.
- You publish a review link and send it to your team or your client.
- Everyone opens the real app in their browser, on a computer or a phone, and clicks a button, a text or an image to attach a comment to it.
- You publish again after your fixes: the same link shows the new version.
For a GitHub file, you can also choose Cloud: the preview then runs online for the whole team, and is published from the web app.
And feedback, all the way to the AI
The most interesting part comes when the loop closes. Riqoch MCP, included in the desktop app, lets your coding agent, Claude Code, Cursor or another, read the review’s comments. Each comment is tied to an element of the app built from your code, which helps the agent find where to make the change. You stay in charge: you are the one who resolves the threads.
We go through the setup in a dedicated article: send review feedback to your coding agent.
In short
Building fast with AI is half the way. The other half is putting the app in front of real people, with feedback precise enough for you, or your agent, to act on. Riqoch takes care of that half, whichever tool wrote the code.