Anthropic ships a commerce agents blueprint on GitHub, with Shopify and Priceline building on it

Anthropic published a reference blueprint for building commerce agents on Claude on 2 September, with the code released as an Apache-2.0 repository at github.com/anthropics/commerce-agents. The repository picked up 1,646 stars in the first three days.

What is in the blueprint

The blueprint has two agents. A shopping agent handles catalog search, multi-item requests, preferences, product comparisons, cart building, and customer service questions inside one conversation. A merchant agent covers sales analytics, inventory tracking, pricing recommendations, and drafting marketing campaigns.

Anthropic says the guardrails are designed to constrain prices and products to the actual catalog data and to avoid manipulative upsell patterns. The examples cover retail, telecom, and entertainment as well as generic commerce, and the code runs on the Claude API, Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI.

Who is building on it

Anthropic names Shopify, Priceline, Visa, Mastercard, Accenture, Intuit, Klaviyo, Wix, Zomato, Fetch, and Square as companies building commerce agents with Claude. Shopify has published a companion repository, Shopify/claude-for-commerce-examples, showing a storefront shopping agent running over the Unified Commerce Protocol with Sign in with Shop, and a merchant agent over the Shopify Admin API.

Anthropic reports that shopping agents built on the blueprint produce carts up to 35 percent larger and shoppers 60 percent more likely to complete a purchase. Those are the company's own figures.

What it means for a team building software

Commerce agents are one of the areas where the boundary between demo code and something a real store can put in front of customers has been unclear. The blueprint moves that boundary. A team that decides to build a shopping or merchant agent no longer starts by inventing a way to bind the agent to a live catalog, or by writing the rules that stop it from promising a price that is not real. Those decisions are made in the reference code and can be argued with rather than started from a blank file.

The named partners also matter for a reader deciding whether to build on top of it. A blueprint that Shopify has published a working example against is one that another Shopify merchant can copy from with lower risk than a pattern that only exists in a research post.

Source

• Building Commerce Agents with Claude, Anthropic, 2 September 2026
• anthropics/commerce-agents on GitHub
• Shopify/claude-for-commerce-examples on GitHub