Claude for Commerce Agents
TL;DR Highlight
Anthropic released a commerce agent blueprint; customers report carts up to 35% larger
Who Should Read
A backend developer adding a conversational shopping agent to a retail, travel, or ticketing service
Core Mechanics
- Shopping and merchant agents included
- The blueprint repo contains a shopping agent and a merchant agent.
- Reference implementations cover retail, travel, telecom, and ticketing platforms.
- A Claude Code plugin is included.
- Build with Messages API or Agent SDK
- Agents can be built with the Messages API, Agent SDK, or Claude Managed Agents (beta).
- They deploy on Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI.
- Shopping agent handles up to checkout
- It includes integration points for catalog, cart, checkout, preferences, and order history.
- Payment is left to your existing checkout or an agentic payments provider.
- It also answers order, return, and refund policy questions in the same chat.
- Prices and products tied to catalog data
- Guardrails constrain prices and products to actual catalog data.
- It is designed to avoid manipulative upsell patterns.
Community Reactions
- Bragging about the numbers feels off
- Some commenters called the 35% larger carts a dark-pattern brag.
- They object to an AI acting like a person to sell more.
- Users want buyer-side agents
- One commenter wants stores to expose real-time catalogs as RAG endpoints for their own agents.
- They argue vendor-controlled agents have adverse interests.
- Odd placement of the agent
- One commenter noted the buying agent sits inside the merchant's app, contrary to the expectation that discovery happens in a separate LLM chat.
- Concerns about examples and safety
- One commenter found the wooden blocks for a 9-year-old example off.
- Another said buyer messages with refund or reprice instructions reaching the context is risky.
- Others questioned why non-deterministic AI belongs in payments.
How to Apply
- Fork the repo and run the demo
- Today, fork github.com/anthropics/commerce-agents and run the shopping agent locally.
- Read how the skills and tools are organized.
- Adapt it to your catalog with Claude Code
- Give Claude Code your catalog, policies, and brand and have it customize the agent.
- Verify prices and products come only from catalog data.
- Add human approval to proposed changes
- Require staff approval before agent-suggested price or promotion changes go live.
- Keep payment in your existing checkout.
Try It
# Paste into Claude Code (run in the folder where you cloned the repo)
git clone https://github.com/anthropics/commerce-agents
Read this repo's shopping agent structure and explain it.
Our catalog: {{catalog format or API}}
Our policies: {{return and refund policy}}
Show a plan to adapt the shopping agent to this catalog first.
Mark the parts where prices and product info come only from catalog data.
Next: run the same request a second time and compare whether you get a different designTerminology
Related Papers
Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy
Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.
Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents
Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.
Harness Engineering 101: The LLM is the brain, the harness is the body
An agent is a loop managing a JSON array; 16 chapters build one from scratch
Harness2 by Google Research – Recursive Agent Harnessing
Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.
Ousted Savannah police officer gave ICE agent husband access to Flock system
A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired
Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling
Running Claude Code agents in tmux works well, but limits appear around four agents.