Anthropic has published a new toolkit designed to help companies stand up conversational shopping and operations assistants on Claude, and payment networks Visa and Mastercard are among the partners helping merchants put those templates into production.
The company framed the release as a practical “blueprint” rather than a finished storefront: reusable patterns, safety constraints, and sample code that engineering teams can adapt in days instead of building agentic commerce from scratch.
The package includes two reference designs.
A customer-facing shopping agent can sit inside a retailer’s site or app, accept requests in everyday language, search a live catalog, remember preferences, compare options, assemble multi-item baskets, display products in the chat, and pass a completed cart to whatever checkout the merchant already uses.
Anthropic is explicit that it is not taking over payments.
Checkout stays with the retailer or with an agentic payments partner.
A second, internal merchant agent is meant for store teams. It can answer questions about what is selling, watch inventory, flag stock that may run out before a promotion, suggest prices and campaigns based on a retailer’s own history, and draft marketing copy.
Any change the internal agent proposes still requires a person to approve it before it goes live.
Anthropic says retailers already running Claude-based shopping agents have seen baskets grow by as much as 35 percent and conversion rise by about 60 percent.
Those figures come from the company and have not been independently audited.
Named early users of Claude-powered shopping experiences include Shopify and Priceline.
Accenture, Visa, and Mastercard are listed as ecosystem partners that will help clients and merchant networks adopt the blueprints.
Shopify has also published public sample code showing how an agent can talk to a live store catalog and cart.
Reference implementations cover retail, travel, telecom, and ticketing, and the same patterns can be customized for other categories.
Developers can deploy the code through the Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI, using the Messages API, Agent SDK, or Claude Managed Agents.
A Claude Code plugin is included so teams can fork the GitHub repository and tailor catalogs, policies, and branding.
Guardrails are meant to keep prices and products tied to real catalog data and to limit pushy upselling.
Live demos and an engineering write-up were released alongside the templates, timed for holiday planning.
The timing matters because card networks have spent more than a year building the trust and token layer that would let a verified agent spend within limits a consumer sets.
Visa’s Intelligent Commerce and Mastercard’s Agent Pay programs issue tokenized credentials and spending rules rather than handing an agent raw card numbers.
Anthropic’s move complements that work instead of competing with it: Claude supplies the shopping brain and operational assistant; Visa, Mastercard, and merchants supply checkout, tokens, and fraud controls.
That division of labor is a deliberate choice.
Anthropic says it does not want to own catalogs, ads, or the last mile of payment.
For retailers, the bet is speed.
Conversational discovery is already common; the harde problem is turning a chat into a trustworthy cart without inventing a new payments stack.
By publishing working agents and partnering with the networks that already move most card volume, Anthropic is trying to make that last step an integration problem rather than a research project. Whether shoppers will actually let an agent finish the purchase at scale remains an open question. The blueprint, the partners, and the holiday calendar are Anthropic’s answer to how quickly the industry can find out.