opencommerce gives your agent one buy tool. The shopper says what they want, the tool searches the stores, shows the products, builds the cart, and places the order once the shopper approves it. Catalogue search, checkout and order tracking sit behind a single endpoint, so you integrate once instead of once per store.

It is an SDK, @agent-cards/commerce-sdk, and a product by Agentcard (opens in a new tab).

Run npx agent-cards@latest init in your app's folder. It signs you in or creates your account, installs the SDK, writes your credentials into your env file and out of git, and builds the buy tool into your code. In an empty folder it sets up a working example agent first.

init needs Node 22 or later. The SDK it installs runs on Node 18 or later and has no dependencies.

The buy tool works with Claude, GPT and Gemini models.

5,026 online stores through one integration, from Amazon, Walmart, Target and DoorDash to flights, hotels and restaurant tables. The buy tool covers four kinds of purchase, and the model picks the right one from the shopper's words: products from online stores, food delivery, a table, or a flight.

Before you send a shopper to a particular store, ask GET /buy/merchants whether Agentcard can pay there. The call names no shopper, so you can check a store before anyone has connected.

Yes. opencommerce by Agentcard lets your agent buy from any online store through one endpoint: catalogue search, cart, checkout and order tracking at 5,026 online stores. You don't build merchant integrations or run an agent browser.

Standard is 1% of each completed order, across every merchant and every payment provider. Enterprise adds API access on top of the tool, on custom fees.

Your agent sends one Vault link. The shopper opens it, saves a card with Face ID or Touch ID, and comes back to the chat. From then on each purchase gets its own approval link, which only the shopper can act on.

You build no sign-up screen and no card form: the Vault link is the sign-up, and your own id for the shopper is their identity in your app. Purchases are paid from Agentcard Vault today, with more wallets coming soon.

No. Every link the turn produces, Vault, approval, store sign-in, product and store, is handed to your own sendLink callback, and what the model reads is the same answer with the links taken out. Your model never reads or writes one.

The shopper, not the model, approves every purchase. Nothing is ordered until they do.

Two ways, depending on how your app is hosted. On a long-running server, call listen once and Agentcard opens the connection to you, so you host no public URL and check no signature. On serverless, mount the SDK's webhook handler on a route instead: it checks each delivery's signature, skips an event it has already handled, and has Agentcard deliver again if your post fails.

Either way you get a message with the text ready to post into that shopper's chat, plus the order events: placed, confirmed, failed and updated.

Under 20 seconds for a catalogue search or a purchase.

We run the purchase flow at each monitored store, search, add to cart and checkout, and publish the result. The monitoring page lists only the cases whose latest test passed, with the last 45 days of history behind each one, so a store that breaks drops off the page instead of sitting there green.

That is the Enterprise plan: API access in addition to the tool, on custom fees. Tell us about your company and we will set up a call.