Claude Commerce Agents: Anthropic’s 2 open source AI agents for ecommerce, one for shoppers and one for merchants

Claude Commerce Agents is a free, open source project from Anthropic with two AI agents for ecommerce: a Shopping Agent that helps customers choose products and a Merchant Agent that helps store operators understand sales and inventory. The code is on GitHub under the Apache 2.0 license. What a company mostly has to build is the connection to its own store.

Everyone is talking about Claude Fable 5.1, but for businesses I think this project is even more interesting.

What does the Shopping Agent do?

The Shopping Agent lives inside the online store, on the customer’s side. The customer types a request in plain language, for example: "I need a laptop under €1,500 that’s powerful enough for video editing". The Shopping Agent searches the catalog, compares products and recommends the best fits.

The Shopping Agent also helps with the cart, orders and store information. Payment stays out of scope: according to Anthropic’s repository, the agent prepares the cart and the store’s own system completes checkout.

What does the Merchant Agent do?

The Merchant Agent is built for the people who run the store. You can ask it, for example, why a product category sold less this week. The Merchant Agent analyzes sales, inventory, catalog, pricing and promotions and helps you understand what’s going on.

Changes proposed by the Merchant Agent don’t go live on their own. Every write stays pending until a person approves it.

What’s inside the Claude Commerce Agents repository?

Anthropic published much more than a demo. The Claude Commerce Agents repository contains almost the entire architecture of both agents:

• the prompts, the core instructions for each agent;
• the skills, the workflows each agent knows how to run;
• the tools, the operations the agent can call;
• the memory, which lets the Shopping Agent remember what the customer told it;
• the guardrails, the checks that limit what the agent can do;
• the components needed to make them run.

The repository includes examples for retail, travel, telecom and entertainment. The agents run on the Messages API, the Claude Agent SDK or Managed Agents.

What does a company need to build to use them?

What’s left to build is mainly the connection to your store. For this, Anthropic defined two interfaces:

• StorefrontBackend for the Shopping Agent: catalog, cart, orders and store policies;
• MerchantBackend for the Merchant Agent: analytics, catalog, inventory, pricing and campaigns.

In practice, these two interfaces define which data and which operations your store must expose to the agent. If you’d rather have it built for you, the integration with catalog, orders and inventory is a custom software development project in its own right.

How do you build the agents with Claude Code?

If you use Claude Code, you can install the commerce-builder plugin included in the repository and ask it to build the agents and connect them to your store’s systems:

claude plugin marketplace add anthropics/commerce-agents
claude plugin install commerce-builder@claude-commerce-agents

The plugin adds commands to Claude Code such as /scaffold-commerce-agent, to generate the agent’s structure, and /review-commerce-agent, to review it.

Is Shopify already using Claude Commerce Agents?

Yes. Shopify published its own implementations of the blueprint on GitHub. There’s a Shopping Agent that works on a live Shopify store and sends the customer to the store’s checkout. And there’s a Merchant Agent that reads products, orders and inventory through the Admin API and applies changes only after an operator approves them.

What does this mean for an ecommerce business?

It means a company can now take an architecture designed by Anthropic, connect it to its own systems and build ecommerce agents without starting from scratch.

There’s one detail to know. Anthropic presents Claude Commerce Agents as a reference implementation: the repository isn’t maintained and doesn’t accept contributions. Whoever adopts it adapts it and then runs it on their own.

If you run an online store, the first step is deciding which data and which operations you’re ready to hand over to an agent.