Inspiration

Today's web apps are built for human clicks and mouse cursors. When autonomous AI agents try to interact with web stores, they're forced to rely on fragile screen-scraping and brittle DOM selectors that break whenever a CSS class changes.

With the emerging W3C Web Model Context Protocol (WebMCP), websites can expose structured, type-safe tools directly to agents inside the browser. We built ShopPulse to show what a real, production-grade agent-native commerce platform looks like.

What it does

ShopPulse unites consumers, merchant operators, and autonomous browser agents in a single, responsive web application:

  • For Shoppers: An AI Co-Pilot uses document.modelContext.configure_bundle to dynamically curate multi-item bundles with instant volume discounts based on natural language requests.
  • For Merchants: A command center with automated purchase order dispatching and econometric price elasticity simulation.
  • Enterprise Safety: A 3-tier permission model requiring explicit human approval on high-risk operations (bulk restocking, financial checkouts).
  • Dual WebMCP Implementation: Supports both imperative JavaScript tool registration (registerTool) and declarative HTML attributes (toolname, tooldescription).

How we built it

  • Framework: Next.js 14 App Router with React and TypeScript.
  • Agent Layer: W3C WebMCP draft specification with structured JSON Schema tools.
  • Visuals: Three.js interactive 3D particle and mesh acceleration.
  • Styling: Tailwind CSS with custom glassmorphism and executive typography.

Challenges we ran into

  • Designing a non-blocking Human-in-the-Loop workflow that safely pauses async tool execution while preserving the browser session.
  • Seamlessly blending agent actions with live UI state updates without reloads.

What's next for ShopPulse

  • Direct integration with Chrome's built-in on-device Gemini Nano via window.ai.
  • WebMCP-native multi-store checkout federation.

Built With

Share this project:

Updates

Submission history