PayPal AI Hackathon demo
Many small makers. One big order. One PayPal checkout.
OpenStall Collective turns home makers who only use WhatsApp into one supplier that AI shopping agents can order from in bulk. A maker sends photos and a voice note, and AI turns them into a listing any agent can read. When an order is too big for one maker, plain code splits it across several, one PayPal approval reserves the money, and each maker is paid the moment she sends a photo of her shipping receipt.
- No real money moves
- PayPal sandbox
- Open source, MIT
Right now on OpenStall
Loading the numbers…
From a voice note to a paid order
How it works
A maker lists by WhatsApp
She sends photos and a voice note in her own language. AI drafts her listing, asks one question at a time for what it misses, and publishes it only when she says yes.
Try it as a sellerA buyer's AI agent asks
Through OpenStall's MCP storefront it searches, reads each maker's trust record, and asks for a quote: 40 cushion covers, at most $30 each, in 30 days.
See the requestSeller agents negotiate within her limits
Each maker's agent offers a price and a line to the buyer. Code checks every price against her floor: no words from a buyer can talk it lower.
See the offersPlain code splits the order
The cheapest valid offers fill it across makers, and the split is explained in plain words. A person approves one PayPal checkout, which only reserves the money.
See the splitEach maker is paid on proof
She photographs her courier receipt and her share is captured and paid, with her tracking on the PayPal order. If she cannot deliver, her share goes back and the order heals itself.
See the paymentsTrust and demand build up
Every delivery signs her trust passport again, which any agent can check. Searches that found nothing become tips telling makers what to make next.
Open the dashboard
Five makers, four countries
Meet the makers
Each one sells only on WhatsApp. Together they can fill an order none of them could fill alone.
Model Context Protocol
For AI agents
Connect any MCP client to the storefront: search, products, trust cards, quotes, a draft collective order and an approval link for its human. No tool can move money. Agents without MCP can read the feeds.
Plug an agent in
MCP storefront (Streamable HTTP)
https://openstall-collective-api.onrender.com/mcpPay-on-proof
Built on PayPal
Pay-on-proof on the payouts path, proven step by step on PayPal's real sandbox. The site runs in the sandbox, or with a simulated PayPal in test mode: no real money moves.
- Orders v2, intent AUTHORIZE
- one approval reserves the whole collective order
- Payments
- a partial capture of each maker's share on proof, a final capture on the last, a void of the rest
- Payouts
- each maker paid on her own rail: PayPal, Venmo for a US maker, or her bank through Hyperwallet in Pakistan (simulated)
- Order tracking
- each maker's courier and tracking number on her capture
- Webhooks
- every event verified with PayPal and applied exactly once
- APIMatic's PayPal SDK
- every Orders and Payments call
Every PayPal product, and how far it has gotRead the code on GitHub
Who it is for
Sellers
For example, a home embroiderer
- Uses
- WhatsApp only
- Wants
- Orders, without a website or an app.
Buyer agents
AI agents shopping for a person or a business
- Uses
- Our MCP server
- Wants
- To find, check and buy from sellers they can trust.
The people behind the agents
Whoever pays for the order
- Uses
- An approval page and PayPal
- Wants
- To stay in control of their money.
Demo viewers and judges
Anyone trying OpenStall
- Uses
- The live demo and the seller chat page
- Wants
- To see the whole flow happen, and try it.
Watch one order happen, start to finish.
A buyer's agent asks, makers' agents bid, PayPal holds the money, and each maker is paid when she ships. It takes about a minute.
Run the story on /demo