Anthropic Releases Open Blueprint for Commerce Agents
Anthropic has released anthropics/commerce-agents, an Apache-2.0 reference blueprint for building shopping and merchant agents. The package includes runnable examples for retail, travel, telecom and entertainment, and works locally with Python 3.11+, Node 22 and an ANTHROPIC_API_KEY. Its runtimes can use an Anthropic client through the Claude API, Amazon Bedrock, Microsoft Foundry or Google Cloud Vertex AI.
The shopping agent operates in a merchant’s application. It can search catalogs, handle multi-item requests, compare products, assemble carts, and answer questions about orders and returns. The merchant agent is designed for staff, supporting sales and inventory questions, pricing and promotion suggestions, and campaign drafts. Both use modular skills over catalog, cart, order and policy systems. A Claude Code plugin can scaffold or review an agent.
Anthropic’s engineering guidance favors one agent with skills over intent routers or separate subagents for every domain, arguing that handoffs can lose context and add latency. The blueprint also represents interface components as typed tools, allowing server validation before rendering. It recommends putting volatile data last for prefix-based prompt caching; reported deployment hit rates range from 90% to 99%. Financial actions, writes and identifiers are controlled by code rather than left solely to the model.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.