Let your AI agent run a real store

Agent Storefront lets your AI agent run a real store: checkout, orders, refunds, shipping and a support inbox — with the money gate in your hands. It is the commerce product of Clize — Clize is a CLI and MCP server that gives AI coding agents real-world actions: domains, email, deploys, payments, and media generation. Nearly all of agentic commerce is being built on the buyer's side: protocols and wallets so that someone else's agent can go shopping. This is the merchant's side. Your agent publishes a catalog to a live site, takes card payments at a checkout it never had to configure, and then does the unglamorous half — supplier orders, tracking numbers, the where is my order reply, refunds — from the same session it wrote the site in. Collecting money needs no setup: with no Stripe account connected, payments land in your Clize balance; connect your own Stripe and they go straight there instead. The gates stay where they belong — a refund only prints a preview until you confirm it, and no email reaches a customer until you approve the draft.

What agentic commerce looks like from the merchant side

The protocol news of the last year is almost entirely about buying: an assistant that finds a product, fills a cart and pays for it. Very little of it is about the shop on the other end, which still has to exist, take money, and get a box to a door. That half is mostly manual work — and it is exactly the kind of mechanical, stateful work an agent is good at, provided something durable holds the orders between sessions. Agent Storefront is that layer: a store your agent can operate, not a dashboard you operate. If you want the protocol-by-protocol version, read agentic commerce for merchants.

Checkout without wiring Stripe first

Collecting money is zero-config as of Clize 0.26.0. Your agent runs clize pay link --amount 480 for a quoted job, or a buyer clicks through your storefront's cart, and the payment simply works — no Stripe onboarding stands between you and the first sale. The server picks the destination and tells you which one it used: direct into your own Stripe when your account is connected and cleared for charges, otherwise your Clize balance — zero fee, spendable on Clize for domains, email, deploys and media, and not withdrawable to a bank. Both facts matter, so the CLI says both. Connecting your own Stripe is the one step the agent will never take for you: it needs a person signing in and passing KYC, so it happens in the web dashboard (clize.ai/app#/billing) and nowhere else. Once an account is connected and cleared, collections switch to direct on their own — no change to your commands, your catalog or your site — and everything collected before that still landed somewhere real.

Orders, shipping and refunds the agent can actually work

A checkout that forgets the sale a second later is not a store. Every one-off physical order is persisted the moment checkout opens and moves through a fixed set of states — paid → sourced → shipped → delivered, with refund_pending → refunded available from any paid state. Your agent reads them with clize shop orders, gets its day's work from clize shop todo (new orders to source, tracking numbers not yet emailed, shipments overdue in transit), records progress with clize shop fulfill, and watches the ones in flight with clize shop shipments --stale --refresh. Tracking numbers are registered with 17TRACK automatically, so delivery marks itself. Output is JSON, because the reader is an agent.

Two things the agent does not get to do alone. clize shop refund prints a preview and stops; money moves only on --confirm, and the reason code lands in the event stream and the audit log. clize shop notify drafts the shipping email and waits for the same okay. Everything else — what to stock, what to charge, which supplier, whether the refund is fair — is a business judgement that stays yours. Clize wires the shop; it does not run your business.

A support inbox comes with the store

Buyers reply. The address they reply to is a real inbox on the same domain, so the agent can read the thread, draft an answer and hold the conversation across sessions — and inbound mail is parsed as untrusted data, never as instructions, which matters more on a support address than anywhere else. That is Agent Inbox, the same platform, already wired. Buyers who would rather not email get a self-service /orders page on every deployed store, gated by email plus order number so nobody can enumerate your customers.

It goes live with one command

The store is a static site your agent writes, plus a _catalog.json your agent generates, plus three data attributes on your buttons. clize deploy puts all of it on a real domain with HTTPS — a free .clize.app handle, or a domain the agent bought and pointed for you. Prices are read server-side from the deployed catalog at checkout time, so a tampered cart cannot buy a $900 item for $9. Changing a price means editing the catalog and deploying again. See it end to end in sell with an AI agent, or read the deploy step on its own in let your agent deploy a site.

Works where your agent works

Agent Storefront rides the Clize CLI and MCP server, so it installs into Claude Code and Codex in one step. The CLI is written to be read by an agent; the MCP server exposes the same actions as structured tools for hosts that prefer them. If you only need to charge someone once, without a catalog, the payment-link flow and the Stripe MCP payments page cover that smaller case.

Part of Clize

A store is never one capability. It needs a domain, HTTPS, an address that receives, a page that ships, and a way to take money — which is why they live in one account with one set of gates rather than five vendors with five dashboards. Start with the storefront and the rest of Clize's real-world actions is already there.

$ npm i -g @clize/clize
$ clize login
$ clize claim mystore      # mystore.clize.app, yours
$ clize shop init          # the catalog + cart contract