Skip to content
NewLet your coding agent close the loop

Build what your users actually want

A feedback widget for your site, a public roadmap your users vote on, and a changelog that tells everyone who asked when it ships.

For founders and small teams, one site or 15.

Free for one site. No card. The widget is the button in the corner.

One request going round the loop: asked on a site, sorted in the inbox, shipped on the public roadmap, and the person who asked hears back.

One prompt setup

Copy the prompt into Claude Code, Cursor or Codex. Your agent finds where the widget goes in your project, adds it the right way for your framework and checks it loads.

Your own line comes with your free account. The agent asks you for it.

Or follow the guide for Webflow, Framer, Squarespace, WordPress, Shopify, Google Tag Manager, Next.js and Vite / React in your free account.

For Claude Code, Cursor, Codex and similar agents.

See the prompt
# TASK: add the Luuphub feedback widget to this site

## OBJECTIVE
Make the Luuphub widget (a feedback button, https://luuphub.com) load on every page of this site with the site's own key, and check it works. Up-to-date instructions: https://luuphub.com/developers/docs/widget.md

## FIRST: GET THE SITE'S KEY (do this before editing)
The widget needs this site's publishable key: `lh_site_` followed by 32 letters and digits. It is not a secret (it is visible in page source) and it comes with a free Luuphub account.
1. Ask the user: "Paste your Luuphub site key, or the script line from your Luuphub install page." Use the key exactly as given.
2. If they have no account yet, tell them where it comes from: sign up free at https://luuphub.com/sign-up and add this site's address. The install page then shows the line, and a setup prompt with the key filled in that also checks the install from the live site; with an account, that prompt is the better one to use.
3. Only if they want to go ahead without a key: install with the placeholder `lh_site_YOUR_KEY` exactly, and end by telling them to replace it with their key in every file you changed. Never invent a key, and never reuse a key from another project or from this text.

## BEFORE YOU EDIT
Inspect the project; do not assume its shape.
1. Identify the framework and version (package.json, config files, template engine).
2. Pick the matching route:
   - Next.js: do not paste the snippet. In every root layout (`app/` or `src/app/` layout that renders `<html>`), `import Script from 'next/script'` and render `<Script src="https://w.luuphub.com/loader.js" data-lh-site="lh_site_YOUR_KEY" strategy="afterInteractive" />` inside `<body>` after `{children}`; on the Pages Router, in `pages/_app` after `<Component {...pageProps} />`.
   - Vite single-page app: paste the snippet into `index.html` and every other HTML entry.
   - Anything else: paste the snippet into the one place that writes the document head for every page (root layout, app shell, base template or a single-page app's `index.html`). Plain HTML files each have their own head: paste it into every page.
   - No site code in this project (a site builder such as Webflow, Framer, Squarespace, WordPress or Shopify): stop and point the user to the builder's steps on their Luuphub install page.
3. Search for `data-lh-site` and `luuphub` (skip node_modules and build output). If the widget is already there, fix that copy instead of adding another.
4. Look for a Content-Security-Policy in response headers, host config or a meta tag. If you find none but the host may add headers you cannot see, put the CSP sources below in your report.

## SNIPPET (verbatim apart from the key; do not reformat, minify or re-quote)
```html
<script>!function(w,d){w.luuphub=w.luuphub||function(){(w.luuphub.q=w.luuphub.q||[]).push(arguments)};var s=d.createElement('script');s.async=1;s.src="https://w.luuphub.com/loader.js";s.setAttribute('data-lh-site',"lh_site_YOUR_KEY");d.head.appendChild(s)}(window,document);</script>
```
Replace `lh_site_YOUR_KEY` with the key from FIRST, wherever it appears (the snippet or the Next.js element).

## RULES
- Once per page: in the place every page shares if there is one, otherwise in each page's head.
- If a CSP exists, extend it and change nothing else. The widget needs https://w.luuphub.com in `script-src`; https://luuphub.com, https://w.luuphub.com and https://vercel.com in `connect-src`; `blob: data:` in `img-src`; `blob:` in `media-src`. For each directive: if the policy has it, append these sources; if it lacks it but has `default-src`, add it with `default-src`'s sources plus these; if it has neither, add nothing, because that type is not restricted. If `script-src` uses a nonce, put the page's nonce on the tag you add, the way the project already does; under `'strict-dynamic'` that nonce, not the host, is what lets the loader in. If the policy forbids inline scripts and uses no nonce, use `<script async src="https://w.luuphub.com/loader.js" data-lh-site="lh_site_YOUR_KEY"></script>` instead.
- Keep the key literal, not in an environment variable: it is publishable and safe to commit.
- Do not vendor, bundle, proxy or self-host the loader; load it from its URL so it stays current.
- Do not lazy-load it, defer it behind an interaction, or gate it on a consent wrapper. If the site has one, keep the widget outside it and give the user these facts, not your own reasoning: until opened it sets no cookie and stores no identifier, fetches only its files and the public site config, cookie-less, and sends one boot ping per session carrying the page address, nothing about the visitor; a visitor token is stored only once someone submits feedback or is identified.
- Do not install any npm package. There is none; the snippet is the whole integration.

## CHECK
1. Run the project's build, if it has one; fix only what this change broke. Outside Next.js, the built or served HTML now contains `data-lh-site`.
2. With a real key: `curl -s https://luuphub.com/api/w/<KEY>/config` returns JSON with `"enabled":true`. `{"enabled":false}` means the key is wrong or the site is paused: ask the user to copy it again from their install page.
3. Deploy the way this project deploys (commit first if it deploys from git). Ask the user before pushing to a shared branch or production. Then the user opens the live site once: a Feedback button appears in the bottom corner, and their Luuphub install page shows the widget as live.

## REPORT BACK
Files changed, whether the real key or `lh_site_YOUR_KEY` is in place (and, for the placeholder, every spot to replace), any CSP change, and the result of CHECK. If you stopped early, say at which step and what the user still has to do.
Paste one line instead
<script>!function(w,d){w.luuphub=w.luuphub||function(){(w.luuphub.q=w.luuphub.q||[]).push(arguments)};var s=d.createElement('script');s.async=1;s.src="https://w.luuphub.com/loader.js";s.setAttribute('data-lh-site',"lh_site_…");d.head.appendChild(s)}(window,document);</script>

Your own line comes with your free account, with guides for each platform.

  • 2.5 KBloader, after your page
  • 24languages in the widget
  • Text, voice, videoplus screenshots and files
  • 1 site freefor as long as you like

The problem

Most feedback dies in an inbox

Luuphub brings it back to the person who asked.

Without a loop

Asked three times, in three places. Nobody hears back.

With Luuphub

One request, sorted, shipped, and everyone who asked is told.

NewAI-native, over MCP

Don’t just read feedback. Point your AI agent at it.

Your coding agent reads the top requests, opens a PR, ships it and tells everyone who asked.

  • Reads votes and comments
  • Opens a PR in your repo
  • Ships only on green CI
  • You choose read only, or read and act

Works withClaude CodeCodexCursorVS CodeGemini CLIWindsurf

$ claude mcp add --scope user --transport http luuphub https://luuphub.com/api/mcp/mcp

Signed in to Luuphub. Read and act.

Build the most-voted open request on example.com and ship it.

  1. get_autopilot_queuenext: “Export archived posts too”, 14 votes
  2. get_postthe request, its votes and comments
  3. start_agent_runbuilding it on a branch
  4. append_run_stepPR #212 opened, CI green
  5. complete_agent_runSHIPPED: moved to Done, changelog out

Loop closed. Everyone who asked gets an email.

An example of a coding agent working through Luuphub’s MCP tools.
  1. Feedback landswidget or board
  2. Agent reads itover MCP
  3. Builds, opens a PRClaude Code, Codex, Cursor
  4. Ships iton green CI
  5. Loop closedeveryone who asked hears back

How it works

One request, from asked to shipped

Follow one example request through all four steps.

  1. A feedback widget on every page

    Visitors type, talk or record. No account needed.

    • Screenshots and files
    • 24 languages
    • Loader under 2.5 KB
  2. Every request in one inbox

    Every site, one inbox. On Pro, AI sorts ideas from bugs and flags duplicates. On Free, one key does it.

    • Transcripts on every plan
    • All sites in one view
    • CSV import
  3. Feature requests on a public roadmap

    Votes push the top requests up. You ship them, or your coding agent does.

    • Board, roadmap and changelog
    • In the visitor’s language
    • Voting needs no account
  4. Everyone who asked hears back

    An email, a changelog entry, a Shipped badge. Nobody has to check.

    • One digest email a day
    • Your feedback tab in the widget
    • One-click unsubscribe

What you get

Everything the loop needs

The widget, the inbox, a public roadmap and a changelog, in one product priced by site, never by user.

  • A feedback page for every site

    Board, votes, roadmap, changelog. 24 languages, right to left too.

  • Voice notes and video

    Talk or record the screen. Every clip gets a transcript.

  • Light on your page

    Loader under 2.5 KB, widget under 30 KB, its own Shadow DOM.

  • All your sites, one dashboard

    Free 1 site, Pro 5, Studio 15. One inbox.

  • Knows who is signed in

    Your app passes the user in. No retyped emails.

  • AI triage

    Sorts ideas and bugs, suggests merges, drafts changelogs.

    Pro and Studio
  • REST API and MCP server

    Read and change feedback from code or any MCP client.

    Pro and Studio
  • Webhooks, Slack and Zapier

    New feedback and status changes, where your team works.

    Pro and Studio

Pricing

Start free. Upgrade when it pays for itself.

No card. Cancel anytime. Free forever for one site.

Annual is 2 months free

Free

$0

No card needed

One site, the whole loop. Forever.

  • 1 site
  • 1 seat
  • Voice & video notes, standard transcripts, 60 min a day
  • Widget + boards + roadmap + changelog
  • You sort feedback by hand on Free. Pro adds AI triage and themes.

Studio

$40.83/mo

billed $490/yr, 2 months free

For agencies and portfolio founders. White-label, every site.

  • 15 sites
  • 10 seats
  • 10,000 AI triage credits / mo
  • Voice & video notes, pro transcripts, 600 min a day
  • Widget + boards + roadmap + changelog
  • AI themes + AI changelog drafts
  • REST API + MCP
  • Webhooks, Slack + Zapier
  • White-label (no Luuphub badge)

See the full comparison, every plan feature by feature.

FAQ

Questions

Something else? Write to hi@aina.rs. The person who builds Luuphub reads every message.

Build the right thing this week.

Start freeGo to dashboard

What is a feedback widget?

A button that opens a feedback box on your site. Visitors write without leaving the page, and Luuphub later shows them what happened.

How long does setup take?

Minutes. Paste one line and you are live. AI triage and agents come with Pro.

Do visitors pick idea or bug?

No. They write one message, or leave a voice note or a video. On Pro, AI sorts it as an idea or a bug. On Free, it lands in your inbox and you sort it.

Do my users need accounts?

No. Anyone can send feedback and vote. Your app can pass signed-in users to the widget.

Will it slow my site down?

No. The loader is under 2.5 KB, waits for your page and never touches your styles.

Can I use it on more than one website?

Yes. Free covers 1 site, Pro 5, Studio 15, in one dashboard.

Can I bring feedback I already have?

Yes. Import a CSV or use the API. On Pro, AI sorts it and flags duplicates.

Do I need a credit card to start?

No. Free runs one site with no card and no time limit. Upgrading starts billing right away.

Stop guessing what to build

The feedback widget that closes the loop.

Free for one site, live in minutes. Pro adds AI triage, more sites and your coding agent.