Anthropic Released Claude Commerce Agents: An Apache-2.0 Blueprint for Shopping and Merchant Agents Across Retail, Travel, Telecom and Entertainment

Details

Anthropic has open-sourced anthropics/commerce-agents, a production-ready reference architecture for building shopping and merchant agents. The repository ships two agents — a shopping agent and a merchant agent — with four pre-built verticals: retail, travel, telecom, and entertainment. Licensed under Apache 2.0, it runs on Python 3.11+ and Node 22, and the same code deploys on the Claude API, AWS Bedrock, Azure Foundry, or Google Cloud Vertex AI. The shopping agent handles catalog search, multi-item comparison, cart management, and customer care within a single conversational session. Its five skill areas are search-discovery, purchase-research, planning-goals, customer-care, and memory-personalization. The merchant agent covers sales analytics, inventory alerts, pricing recommendations, and campaign drafts for store staff. Both run through three deployment modes: the Claude Messages API, the Claude Agent SDK, and Claude Managed Agents (beta). The core architectural argument is the most transferable part: Anthropic advocates skill-based agents over intent routers and per-domain subagents. Each subagent handoff is state-lossy — cart, preferences, and order history live in the orchestrator, and handoffs can multiply token costs and add latency. Skill instructions load directly into the agent that already holds session context. Anthropic reports that in enterprise deployments, single-agent-with-skills outperforms both monolithic prompts and multi-subagent designs on quality, often at lower cost and latency. A Claude Code plugin (commerce-builder) scaffolds new agents or reviews existing ones via /scaffold-commerce-agent and /review-commerce-agent commands — letting teams skip boilerplate and focus on catalog and policy integration.