Anthropic Open-Sources Claude Commerce Agents, With Shopify, Visa, and Mastercard as Early Users

Anthropic published Claude Commerce Agents on September 2, 2026 — an open, Apache-2.0-licensed blueprint for building commerce agents on Claude, described by the company as the harnesses, patterns, and guardrails an engineering team needs to get a commerce agent running in days rather than months. The repository (anthropics/commerce-agents) includes both a shopping agent and a merchant agent, along with four runnable reference implementations spanning retail, travel, telecom, and entertainment.

What's Actually in the Blueprint

• A shopper agent meant for use within a retailer's app or ecommerce site, handling catalogue search, multi-item carts, comparisons, service questions, and checkout via a conversational interface
• A merchant agent facing an organization's own team members, assisting with sales analytics, inventory, pricing recommendations, and campaign drafting with human approval built in
• Deployment options via the Claude API, Amazon Bedrock, Microsoft Foundry, Google Cloud Vertex AI, or as a Claude Code plugin — not locked to a single infrastructure path

Early Adopters and Reported Results

Early users of the blueprint include Shopify, Visa, Mastercard, and Accenture. Anthropic reports that retailers running shopping agents on Claude have seen carts up to 35% larger and shoppers 60% more likely to complete a purchase — figures worth treating as Anthropic's own reported results rather than independently audited numbers, though the specific early-adopter list is a meaningful real-world signal on its own.

Why Open-Sourcing the Blueprint Matters

Rather than shipping commerce-agent capability only as a closed managed product, Anthropic is publishing the actual harness code, prompting patterns, and safety guardrails as a forkable starting point — compressing what previously required months of independent agent-architecture trial and error for any team attempting agentic commerce. For businesses building or evaluating a shopping or merchant-facing AI agent, this is a legitimate reference point to build from or benchmark a vendor's proposal against, rather than reinventing agent architecture decisions from scratch.