Actually Instant

Tell your agent what you need. Actually Instant uses WebMCP to compare stores, products, substitutions, fees and offers, then builds checkout-ready carts for you to review and choose.

Comment

Inspiration

Thinking through the challenge, the first thing that came to mind was a real pain point for me: spending an hour or more shopping for groceries online. It takes time and effort to do it carefully. Otherwise, you end up taking suboptimal (more expensive, less healthy, worse delivery times, etc.) decisions.

Most grocery delivery services let you create multiple carts at different stores, but you don't have an easy way to compare prices including promos, discounts, store-specific offers, payment discounts, delivery times, etc.; you have to move back and forth between stores no matter what (some of them have some sort of shy AI integrations already though).

Why not give your AI agent a chance to do it faster and better? Obvious choice!

What it does

You tell your agent/assistant what you need. Casually. It then compares products across multiple stores and creates the best cart for you to simply pay and move on with life.

How we built it

Existing grocery delivery and pick-up services do a great job, so for this demo I created a fake (parodying a famous service) but realistic enough grocery shopping web app to make the experience feel close to the real thing.

Built with: Vite, React (SPA), shadcn/ui, WebMCP (imperative API), Cloudflare Pages and R2; The React interface and WebMCP tools are adapters over the same application service:

  • context/shop-context.tsx connects React with the shared service using useSyncExternalStore.
  • webmcp/webmcp-tools.tsx registers WebMCP tools and calls the same service methods.

Challenges we ran into

I found out about the challenge a few days ago and didn't have much time for it, so I started quite late (yesterday; 09/02). By chance the deadline was extended due to a network outage that affected Codex during the day. That gave me time to finish the video properly.

Accomplishments that we're proud of

I've had so much fun trying to make the experience feel realistic. The end result looks really nice. The promo video is quite cool too.

I tried really hard not to fall into AI-slop-vibes territory. I think I did well overall.

What we learned

Further confirmation on how much more useful AI agents become when we give them enough (and the right) tools. And how easy is to integrate WebMCP into web clients!

What's next for Actually Instant

Nothing really. It's just a demo! Hopefully the real services integrate WebMCP into their apps soon! Looking forward to it.

Built With

Share this project:

Updates

Submission history