Claude's commerce blueprint ships two reference agents

Anthropic has released a commerce blueprint for Claude, citing retailers whose shopping agents on Claude produce carts up to 35% larger and shoppers 60% more likely to complete a purchase. The package shipped on September 2 with reference implementations, guardrails and a Claude Code plugin, according to the Claude blog.

At a glance

• The repository ships two working reference implementations, a shopping agent that sits inside a retailer's app and a merchant agent for store operators, with variants for retail, travel, telecom and ticketing platforms.
• Both agents can be built on the Messages API, the Agent SDK or Claude Managed Agents, now in beta, and deploy through the Claude API, Amazon Bedrock, Microsoft Foundry or Google Cloud Vertex AI.
• Guardrails constrain prices and products to actual catalog data and rule out manipulative upsell patterns, while merchant-side changes proposed by the agent require a human approval before anything goes live.

Agent scaffolding is where most commerce teams lose weeks: tool iteration limits, prompt caching, in-conversation UI, the plumbing between a catalog and a checkout. Shipping that as a forkable repository with a Claude Code plugin reads as an attempt to make Claude the default substrate for agentic shopping ahead of the holiday season, rather than one option among several. The conversion figures are the sales argument; the harness is the retention one.

The shopping agent covers catalog, cart and checkout but leaves payment to the retailer

The shopping agent lives inside a retailer's own app or website. The blueprint provides integration points for catalog, cart, checkout, customer preferences and order history, and deliberately leaves payment out, whether that is an existing checkout flow or an agentic payments provider.

Given a request such as a tent, sleeping bag and stove for a weekend trip with two children, the agent searches the catalog, assembles the multi-item set, tailors suggestions to remembered preferences and renders products, comparisons and the cart inside the conversation rather than as plain text. Customer service questions about order status, returns and refund policy stay in the same thread.

The merchant agent recommends discounts and waits for a person to approve them

A user can ask what should be discounted to clear last season's inventory and get an answer built on the store's own data. The merchant agent answers questions about sales performance, tracks inventory and flags problems such as an item about to sell out before a promotion starts, recommends pricing and promotions from the store's sales history, and drafts marketing campaigns to move stock.

In the repository, those capabilities ship as skills for sales analytics, catalog and inventory management, marketing and promotions, and in-portal UI including charts and dashboards. When the agent proposes a change on its own, a person approves it before anything goes live.

Accenture research finds 85% of consumers open to collaborating with an AI agent

Shopify, Priceline and other enterprise customers already run agents on Claude that let shoppers search in plain language, compare options and buy. Anthropic is also working with solutions and ecosystem partners including Accenture, Mastercard and Visa to bring the blueprint to clients and merchant communities.

Accenture's Kath Gramling, who leads the firm's consumer goods, retail and travel practice, cites research finding 85% of consumers now open to collaborating with an AI agent and nearly three in four who would trust a personal AI agent more than their best friend to make a purchase for them.

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

What the blueprint does not cover

The code is available now and payments remain out of scope: the shopping agent hands off to an existing checkout or an agentic payments provider, and no pricing or licensing terms are given for the code itself. The repository sits at github.com/anthropics/commerce-agents, with self-guided vertical demos, an engineering deep-dive on how the agents were built and a webinar walkthrough.