Claude Commerce Agents Operational

Claude Commerce Agents is Anthropic's open-source commerce agent project, featuring shopping and merchant agents. Its tools cover four industries: retail, travel, telecommunications, and ticketing, with built-in multi-layer safety guardrails, including temporary staging of all writes for approval and keeping the model away from payment processing. In testing, Claude Commerce Agents increased shopping cart size by 35% and conversion rates by 60%. The tools support three deployment options: Messages API, Agent SDK, and Managed Agents, and can connect to platforms such as AWS Bedrock and Google Cloud.

Follow WeChat and reply “开源” to join the AI open-source project community.
Shopping agent: Embed it in a merchant app or website to help consumers search using natural language, compare prices, plan shopping lists, and build shopping carts.
Merchant agent: Gives operations teams capabilities for sales analysis, inventory alerts, pricing and promotion recommendations, and marketing campaign drafting.
Multi-industry coverage: Provides runnable reference implementations for four vertical industries: retail, travel, telecommunications, and ticketing.
Safety guardrails: Prices and inventory are strictly anchored to real catalog data; all merchant-side writes are staged for human approval, and the model does not access payment processes.
Three deployment options: Supports Messages API, Agent SDK, and Managed Agents.
Claude Code plugin: Automatically generates project code, business workflows, and test suites through a scaffolding command, enabling agents to be built within hours.
Memory and personalization: Remembers user preferences across sessions and automatically extracts and updates user profiles to tailor recommendations.
In-conversation UI: Product cards, comparison views, and shopping cart status are rendered directly in the conversation interface.

Fork the repository: Clone anthropics/commerce-agents locally and install the dependencies for Python 3.11+ and Node 22.
Configure credentials: Copy .env.example to .env and enter ANTHROPIC_API_KEY.
Run an example: Execute python scripts/run_demo.py retail to launch interactive demos for the four industries, including retail.
Connect your backend: Implement the StorefrontBackend or MerchantBackend interface to connect the agent to your catalog, shopping cart, inventory, and order systems.
Claude Code scaffolding: After installing the plugin, run /scaffold-commerce-agent to automatically generate project code adapted to your technology stack.
Choose a deployment method: Use Messages API for a custom loop, Agent SDK for a quick start, or Managed Agents for hosted deployment.
Toggle features: Use enable_* configuration options to disable unneeded capabilities, such as disabling related tools and prompts when there is no shopping cart.
Customize Skills: Create a directory with SKILL.md under skills/ to extend workflows for product discovery, customer service, and other business functions.
Verify and test: Run pytest and scripts/smoke_chat.py to ensure the agent behaves as expected.

Ready to use and fast to deploy: Provides a complete repository that can be forked directly, enabling engineering teams to connect shopping or merchant agents to their own systems and launch within days.
Flexible deployment with multiple paths: Supports Messages API, Agent SDK, and Managed Agents, with seamless integration with mainstream cloud infrastructure such as AWS Bedrock and Google Cloud.
More productive development with plugin acceleration: The Claude Code plugin uses a scaffolding command to reduce agent development from days to hours and automatically generates test suites based on your own catalog and rules.
Simple architecture without redundancy: Uses a “single agent + Skills” architecture, avoiding the state handoff overhead and additional latency of multi-agent orchestration in strongly contextual, continuous scenarios such as shopping.
Built-in security enforced in code: Fences, source gating, permission ceilings, and approval gates are enforced at the code level, ensuring security without relying on model self-restraint.
Payment isolation eliminates risk: The model only builds the shopping cart; checkout URLs and payment operations are handled entirely by the merchant system, and the model cannot see any payment information.
Staged changes with human oversight: All merchant-side writes are staged by default and take effect only after human approval, preventing AI errors from directly affecting live business operations.

Project website: https://claude.com/solutions/commerce
GitHub repository: https://github.com/anthropics/commerce-agents

Intelligent shopping assistance for e-commerce platforms: Consumers describe their needs in natural language, and the agent automatically searches, compares prices, plans, and builds a shopping cart.
Merchant operations decision support: Operations teams ask about sales performance and inventory status; the agent proactively alerts them to slow-moving or out-of-stock risks and recommends pricing and promotion plans based on historical data.
Travel itinerary planning: Users provide multi-dimensional travel requirements, and the agent combines price comparisons, scheduling, and value analysis in a single conversation to generate a complete itinerary.
Intelligent telecommunications plan recommendations: Based on the user's account usage, budget, and number of lines, the agent analyzes plan options, discloses regulatory fees, and recommends the optimal contract combination.
Ticketing and event services: Handles limited-time reservations, waitlists, ticket transfers, and venue seating-map navigation for concerts and sporting events while ensuring transparent fees and real-time inventory synchronization.