Is the Claude commerce agents blueprint free to use?

Yes. The commerce-agents repository is published under Apache-2.0, so the code can be used and modified commercially. Anthropic labels it a reference implementation that is not maintained and does not accept contributions, so treat it as a starting point you will own. Running the agents still costs whatever the underlying Claude API or cloud provider usage costs.

Can the shopping agent take a payment?

No. The commerce-agents blueprint places no orders and charges no cards. Checkout is rendered for the host application to complete, which keeps payment handling, fraud rules and credentials outside the agent. Anthropic's safety document lists authentication, authorization, rate limiting, business and fraud rules, payment processing, memory retention and log hygiene as things a deployment must own itself.

What stops the merchant agent from changing prices on its own?

Guardrails run both when a change is staged and again at apply time, checking items per change, price-move limits, promotion depth, restock size, campaign budget and protected fields. With require_host_approval enabled, apply_change only succeeds for IDs the host has marked approved, and only listing or campaign IDs that came back from a tool call can be staged at all.

Which verticals does the blueprint actually implement?

Four, each with both a storefront and a merchant portal: retail (search, comparison, cart, checkout), travel (date-bound inventory and occupancy), telecom (account context, a plan matrix and regulated fees), and entertainment or ticketing (timed holds, waitlists and venue maps). Anthropic says Accenture, Mastercard and Visa are working with it to help teams put the blueprint into production.

Anthropic · 2026-09-02 · major

Claude commerce agents — Anthropic's blueprint for shopping and merchant bots

Anthropic published an open reference blueprint for building commerce agents on Claude. The Apache-2.0 repo ships a shopping agent, a merchant agent, four storefronts, and guardrails that stage every merchant write for human approval.

GitHub page for anthropics/commerce-agents, Anthropic's commerce agent blueprint — GitHub

Anthropic's commerce-agents repo is a reference build showing how shopping and merchant agents should be wired, fenced and approved.

Key specs

GitHub stars	1,924

Quick facts

Maker	Anthropic
License	Apache-2.0
What it is	Reference blueprint, not a maintained product
Reference agents	Shopping agent and merchant agent
Example verticals	Retail, travel, telecom, ticketing
Requirements	Python 3.11+ and Node 22
Runs on	Claude API, Amazon Bedrock, Microsoft Foundry, Google Vertex AI

What is it?

Two working reference agents are the point of Claude for commerce agents, the blueprint Anthropic published on September 2, 2026. A shopping agent searches a catalogue, compares products, builds a cart and answers policy questions; a merchant agent reads sales performance, tracks stock and drafts pricing, promotion and campaign changes. Everything runs against fictional stores, and no order is ever placed or charged.

How does it work?

The repo splits into seven pip packages — a shared commerce-common plus core, runtime and SDK layers for each agent — so agent logic stays separate from transport. Three execution paths produce the same behaviour: a hand-written loop on the Messages API, the Claude Agent SDK managing the loop, and Managed Agents calling your own MCP server. After cloning, python scripts/run_demo.py retail brings up an API on port 8000 and a storefront on port 3000.

Why does it matter?

Safety in a buying agent is where teams usually slip, and this blueprint enforces it in code rather than in the prompt. Third-party text is sanitized and wrapped in fixed-label fences, only product IDs returned by a catalogue tool can enter a cart, checkout is handed back to the host application so nothing charges a card, and a merchant change only applies to IDs a human has approved. Anthropic says retailers running shopping agents on Claude have seen carts up to 35% larger and shoppers 60% more likely to finish a purchase.

Who is it for?

engineering teams building retail or marketplace agents

Frequently asked questions

Is the Claude commerce agents blueprint free to use?: Yes. The commerce-agents repository is published under Apache-2.0, so the code can be used and modified commercially. Anthropic labels it a reference implementation that is not maintained and does not accept contributions, so treat it as a starting point you will own. Running the agents still costs whatever the underlying Claude API or cloud provider usage costs.
Can the shopping agent take a payment?: No. The commerce-agents blueprint places no orders and charges no cards. Checkout is rendered for the host application to complete, which keeps payment handling, fraud rules and credentials outside the agent. Anthropic's safety document lists authentication, authorization, rate limiting, business and fraud rules, payment processing, memory retention and log hygiene as things a deployment must own itself.
What stops the merchant agent from changing prices on its own?: Guardrails run both when a change is staged and again at apply time, checking items per change, price-move limits, promotion depth, restock size, campaign budget and protected fields. With require_host_approval enabled, apply_change only succeeds for IDs the host has marked approved, and only listing or campaign IDs that came back from a tool call can be staged at all.
Which verticals does the blueprint actually implement?: Four, each with both a storefront and a merchant portal: retail (search, comparison, cart, checkout), travel (date-bound inventory and occupancy), telecom (account context, a plan matrix and regulated fees), and entertainment or ticketing (timed holds, waitlists and venue maps). Anthropic says Accenture, Mastercard and Visa are working with it to help teams put the blueprint into production.

Try it

git clone https://github.com/anthropics/commerce-agents