Shopping and merchant agent templates land on GitHub

Anthropic published a blueprint for building commerce agents on Claude, and says retailers already running shopping agents on its models see carts up to 35% larger and shoppers 60% more likely to complete a purchase. The release, dated September 2, includes reference implementations of a shopping agent and a merchant agent, according to the Claude blog.

At a glance

• Anthropic ships the harnesses, patterns and guardrails as a public repository with a Claude Code plugin, aiming to put a working commerce agent in front of an engineering team within days.
• The merchant agent answers questions about sales performance, tracks inventory, recommends pricing and promotions from a store's own history, and drafts campaigns, with a person approving any proactive change before it goes live.
• Code deploys on the Claude API, Amazon Bedrock, Microsoft Foundry and Google Cloud Vertex AI, and Anthropic names Accenture, Mastercard and Visa as partners bringing the blueprint to merchant communities.

The gap the blueprint targets is scaffolding rather than model capability: teams building commerce agents rebuild the same catalog, cart and checkout plumbing each time. Anthropic appears to be standardizing that layer at the moment payment networks and platforms are defining their own agentic commerce rails, which makes a reference implementation as much a distribution play as a developer convenience. Holiday planning cycles set the timing.

The shopping agent handles catalog, cart and checkout, and leaves payment to the merchant

The shopping agent sits inside a retailer's app or site. The blueprint provides integration points for catalog, cart, checkout, customer preferences and order history, and leaves payment to the merchant, whether that means an existing checkout flow or an agentic payments provider.

Anthropic's example request is a customer asking for a tent, sleeping bag and stove for a weekend trip with two children. The agent searches the catalog, assembles multi-item orders, remembers preferences, renders products, comparisons and the cart inside the conversation, and answers order-status, return and refund questions without routing the shopper to a support page.

In the repository, these capabilities ship as skills and tools covering catalog search, multi-item planning, deep research, personalization, customer care and in-conversation UI. The guardrails are designed to constrain prices and products to actual catalog data and to avoid manipulative upsell patterns.

The merchant agent ships as skills for analytics, inventory and promotions

The merchant agent is aimed at the people running the store. Anthropic's example question is what to discount to clear last season's inventory, answered from the retailer's own data, and the agent can flag an item likely to sell out before a promotion starts.

Those capabilities ship as skills for sales analytics, catalog and inventory management, marketing and promotions, and in-portal UI such as charts and dashboards. When the agent proposes a change on its own, Anthropic's design requires a person to approve it before anything reaches customers.

Anthropic offers a self-guided demo of both agents before any code is written, with live demos for each of the four verticals covered: retail, travel, telecom and ticketing. The stated target is an engineering team getting a commerce agent running in days.

Wix says its engineers had an agent taking prompts in fifteen minutes

Wix says its engineers had a working commerce agent taking prompts within fifteen minutes, according to Dror Zalika, head of commerce. Zomato's Akhil Bansal, a senior engineering manager, says the blueprint ran with no blockers and bakes in practices his team recognized, including tool iteration limits and prompt caching.

Shopify is building a reference storefront implementation on the blueprint, connecting it to a merchant's store through Catalog, UCP and Shop Sign-in, says VP of product Vanessa Lee. Accenture's latest research found 85% are open to collaboration with an AI agent, said Kath Gramling, its global consumer goods, retail and travel lead.

Priceline is named among the enterprise customers, alongside Shopify. Cobus Kok, VP of AI experiences, says the latest generation of the Penny travel assistant was built on Claude for reasoning across flights, hotels and cars. Square, Intuit and Klaviyo also appear among the companies quoted.

Where the code and demos sit

The repository is on GitHub as anthropics/commerce-agents, with an engineering deep-dive published alongside it and per-vertical demos on Anthropic's commerce solutions page. A webinar with live walkthroughs is open for registration, though no date for it is given in the announcement. Anthropic also points teams to its sales team for implementation discussions and working sessions.