Anthropic's official account @ClaudeDevs announced on X: "We're open-sourcing Claude Commerce Agents. This is a blueprint for building shopping and merchant agents, with reference implementations across retail, travel, telecom, and entertainment."
What Was Announced
Claude Commerce Agents is a blueprint — a design framework for developers — for building shopping agents and merchant agents. A shopping agent acts on behalf of a consumer to assist with purchases, while a merchant agent supports seller-side operations. Anthropic is open-sourcing this blueprint along with reference implementations, meaning working example code, across retail, travel, telecom, and entertainment.
What It Means for Users
Developers and businesses can now use this blueprint as a starting point to build shopping and commerce-related agents tailored to their own services. Because reference implementations are provided across several distinct industries, teams can adapt the examples closest to their sector rather than starting from scratch.
How to Use It — Three Starting Points
According to the official README (anthropics/commerce-agents), there are three ways to get started. You can run the demos without writing any code, or let the bundled Claude Code plugin scaffold an agent against your own systems.
- ① Run the demos — Launch the four vertical examples (retail, travel, telecom, entertainment) locally as-is. Requirements: Python 3.11+, Node 22, and an Anthropic API key.
- ② Build your own with the Claude Code plugin — The bundled
commerce-builderplugin asks about your stack, plays the plan back, and builds the project. - ③ Choose a runtime — The same prompt, skills, and tool contracts run on the Messages API, the Claude Agent SDK, or Claude Managed Agents (beta).
① Run the demos — all four verticals locally
git clone https://github.com/anthropics/commerce-agents.git && cd commerce-agents
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt # the seven packages and their pinned dependencies
cp .env.example .env # add ANTHROPIC_API_KEY
(cd examples && npm ci) # the eight web apps share one workspace
python scripts/run_demo.py retail # API :8000 + storefront :3000Replace retail with travel, telecom, or entertainment for the other verticals. --merchant starts the merchant portal instead of the storefront, and --all starts both. Storefronts use ports 3000–3003 and portals 3100–3103. Each example's README has a "Try" section suggesting prompts to test.
② Scaffold an agent for your own service with the Claude Code plugin
With the repo cloned (the plugin reads it as the reference implementation), run:
claude plugin marketplace add anthropics/commerce-agents
claude plugin install commerce-builder@claude-commerce-agents
claude
/scaffold-commerce-agent a shopping assistant for our storeThe plugin provides four commands.
- /scaffold-commerce-agent — Asks about your stack, confirms the plan, and builds the project.
- /add-commerce-flow — Adds a new flow to a generated agent.
- /author-commerce-evals — Writes evals to verify agent quality.
- /review-commerce-agent — Reviews an agent you already have against the blueprint.
Per the README, each command also runs automatically when a request matches its description, so naming it is optional. If plugin installation is new to you, see our Claude plugins guide first.
③ Runtimes — Messages API, Agent SDK, Managed Agents
All three paths share the same prompt, skills, and tool contracts.
- Messages API — The reference loop. Pass your backend and skills directory to
ShoppingAgentand stream turns withstream_turn. - Agent SDK — The same definition with the SDK running the loop, e.g.
python shopping-agent/runtime-agent-sdk/main.py --once "a two-person tent under $250". New to the SDK? See our Claude Agent SDK guide. - Managed Agents (beta) — A hosted agent that calls your MCP server, deployed with
scripts/deploy_managed_agent.sh.
Beyond the Claude API, the announcement and the repo's docs/deployment.md state that the code also deploys on Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI.
What the Two Agents Actually Do
- Shopping agent — Catalog search, product comparison, multi-item planning, cart building and checkout handoff, order status and return/refund policy answers, and remembering customer preferences. A deployment implements the
StorefrontBackendinterface over its catalog, cart, order, and policy systems. Its five flows are defined as skills (SKILL.md) undershopping-agent/skills/. - Merchant agent — Explains sales performance, maintains listings, acts on inventory and order alerts, recommends pricing and promotions, and drafts campaigns. A deployment implements
MerchantBackendover its analytics, catalog, inventory, pricing, and campaign systems. Every write stays staged until a person approves it.
To understand the skill structure, our Claude Code skills guide covers the SKILL.md format.
Safety Design to Know Before You Use It
The README is explicit that this repo is not a finished product to deploy as-is.
- Everything is fictional — Every company, brand, product, and person is made up; the only company is ACME.
- No real transactions — Nothing places an order, charges a card, or changes a live listing on any path.
checkoutrenders the cart for the host to complete, and payment is left to your existing checkout or a separate payments provider. - Catalog-grounded guardrails — Per the announcement, the agent is designed to constrain prices and products to actual catalog data and to avoid manipulative upsell patterns.
- Production security is on you — The examples have no authentication and the MCP servers bind to loopback. Business rules, authorization, and compliance belong to the deployment.
- No MCP connectors ship — Both agents reach your systems through the backend interfaces. Where an official connector exists for analytics warehouses, payments, or delivery tools, the README points to it as the integration target.
License, Maintenance, and Resources
- License — Apache License 2.0 (Copyright 2026 Anthropic PBC), a permissive license allowing commercial use, modification, and redistribution.
- Maintenance — The README states it is a reference implementation that is not maintained and does not accept contributions. The expectation is that you fork and manage it yourself.
- Announced figures — Anthropic says retailers running shopping agents on Claude have seen carts up to 35% larger and shoppers 60% more likely to complete a purchase. These are Anthropic's figures from its own customer cases; no independent verification was found.
- Partner implementation — Shopify separately published an example repo (
Shopify/claude-for-commerce-examples) that connects the blueprint to a real Shopify store.
Official resources: GitHub repo · Announcement · Engineering deep-dive · Live vertical demos
What Remains Unclear
The announcement mentions a webinar, but its schedule must be checked on the registration page. The performance figures above come from Anthropic's customer cases and may differ in your environment. Since the repo is explicitly unmaintained, compatibility with future Claude model and SDK changes is something you will need to verify yourself.