Hold My Hand combines a lightweight help widget with MCP-fed product context so your users get reliable, single-step answers that point to the right control and explain the rest of the path when a menu is involved.
Only UI structure is sent — never screenshots, field values, or user data.
See it work right now — click the help button (bottom-right) and ask:
Hold My Hand will spotlight the next thing to click and walk you through the path.
Drop the prompt below into your AI coding tool. It connects to the MCP endpoint, generates the product context file, and installs the widget script for you.
Generating your 10-day trial key…
Works in Lovable, Cursor, Windsurf, and Claude Code. It teaches the product to your AI, creates the context file, and installs the widget in one pass.
Set up Hold My Hand in this app. 1. Fetch the spec from this MCP endpoint: https://xnmyrtavdhgoubewxcvi.supabase.co/functions/v1/get-instrumentation-spec 2. Follow it to generate public/holdmyhand.map.json describing: - what this app does (app.overview, app.audience) - domain terms (glossary) - common user tasks step-by-step (tasks) 3. Add this script to index.html <head>: <script async src="https://xnmyrtavdhgoubewxcvi.supabase.co/functions/v1/sdk" data-key="GENERATING..."></script> API key (trial — expires in 10 days, 25 questions): GENERATING... Do not add any other attributes or imports.
<head>Skip the MCP and context file if you want the fastest install. The widget still maps your UI at runtime and gives single-step guidance, just with less product-specific detail.
<script async src="https://xnmyrtavdhgoubewxcvi.supabase.co/functions/v1/sdk" data-key="GENERATING..."></script>
Your users click it, ask anything, and Hold My Hand spotlights the right element with one answer that also explains the menu path when needed.
The widget only activates when a user clicks the help button — nothing runs in the background.
When a question is asked, it sends the AI a list of the button and link labels currently visible on the page (things like "Settings", "Export", "Invite team"). It also sends the product description you set up during install.
It does not send: screenshots, page content, form data, or anything a user has typed. Password fields send the label "Password" — not the value.
Think of it as a table of contents for your UI, not a snapshot of your screen.
Want to connect the MCP server to Lovable, Cursor, or Claude, or wire it to your custom domain? Sign in to see the per-key MCP URL and step-by-step instructions in your dashboard.
When a user clicks the help button, Hold My Hand reads the question, scans the current page, consults the context your AI generated through MCP, and then highlights the best next control with a concise answer.
Paste one prompt into Lovable, Cursor, or Claude. Your AI explores the app and writes the product context file through MCP.
A subtle help button appears in the corner. Users ask in plain English, with no commands or training required.
Hold My Hand pins the best next element on screen and explains the rest of the path when the answer lives inside a menu or submenu.