By 
- Shopify built a working AI shopping assistant on top of Anthropic’s open-source commerce-agents blueprint.
- It reaches a store’s real catalog, cart, and checkout through Shopify’s Universal Commerce Protocol (UCP), exposed via MCP.
- The assistant builds carts and answers questions, but hands buyers to Shopify’s own checkout for payment — it never completes the transaction itself.
- A companion merchant agent proposes inventory and pricing changes; nothing is applied without human approval.
- Merchants can get a no-code version through Shopify Inbox, or deploy the open-source build for full customization.
THE CONTEXT
What Shopify Actually Shipped
Merchants can now deploy a working shopping assistant that searches live inventory, builds carts, and hands buyers to checkout, without touching payments itself.
Shopify published a working reference implementation on GitHub, called Shopify-claude-for-commerce-examples, built on Anthropic’s commerce-agents blueprint. It ships two separate agents that work on opposite sides of a store
The storefront agent. Pointed at any Shopify store’s domain, this customer-facing assistant brands itself automatically with that store’s name, logo, colors, and best sellers. From there it searches the live catalog, builds a real cart across a multi-turn conversation, and answers policy or FAQ questions pulled from the store’s own content.
When a customer is ready to buy, it hands them to the store’s existing checkout page. The checkout, shipping, and payment stay entirely on Shopify’s own infrastructure. An optional “Sign in with Shop” flow unlocks personalized results for returning customers. While keeping login tokens server-side, out of reach of the model, the browser, or application logs.
The merchant agent.This operator-facing agent uses Shopify’s Admin API to read products, orders, and inventory, proposing changes like pricing adjustments and restocking. It requires human approval for any action, with all proposals logged. Shopify provides this feature as a backend only, with no prebuilt dashboard for approvals.
HOW IT CONNECT
How It Connects Under the Hood
The technical chain behind both agents involves three distinct layers, often conflated in coverage of the release:
UCP → MCP → Claude blueprint → Shopify’s build
01
UCP (Universal Commerce Protocol)
is Shopify’s own standard describing what a store can do programmatically — search a catalog, build and modify a cart, convert it into a checkout, and monitor order status.
02
MCP (Model Context Protocol)
Shopify exposes these UCP functions through MCP (Model Context Protocol) servers, the open connector standard that lets any compatible AI agent call a service’s functions in a consistent format.
03
Anthropic’s commerce-agents blueprint,
released September 2, 2026, supplies the reasoning layer, a reference design for how an agent should interpret a shopper’s request and decide what to ask a connected store for.
04
Shopify’s new repository
is the integration layer wiring all three together against a live store.
In short, UCP defines what the store can do, MCP is the channel the AI uses to reach it, and Anthropic’s blueprint is the logic deciding what to ask for. None of the three is new by itself , what’s new is a public, working example of all three connected to a live Shopify store.
AVAILABILITY
Who Can Use This
01
No-code path
Merchants who want a managed option with no development work can set one up through Shopify Inbox.
02
Developer path
Developers who want a fully custom, branded build can deploy the open-source code directly. This requires setting up a Python and Node environment, API credentials, and hosting — it’s not a one-click merchant-admin feature.
03
The platform limit
Because the implementation depends on Shopify’s own UCP and MCP servers, it’s currently specific to Shopify stores. A merchant on a different e-commerce platform would need that platform to build an equivalent connector before something like this would work for them
THE NUMBERS
The Numbers, With a Caveat
Anthropic has said carts built through agents like this run up to 35% larger, with shoppers 60% more likely to complete a purchase. Those figures come from Anthropic’s own reporting rather than an independent benchmark, and should be treated as a company claim pending outside verification.