Claude for Commerce explained: what Anthropic shipped

On 2 September 2026 Anthropic published a blueprint for building commerce agents on Claude and opened the code under Apache 2.0. It is not a shopping app inside Claude, and it does not take payment. Here is what shipped, who is named, what the headline numbers describe, and what is missing.

Claude for Commerce is Anthropic's blueprint for building commerce agents on Claude, published 2 September 2026 as open-source reference code. A merchant runs it inside its own app: a shopping agent, a merchant agent, and payment left to the merchant (Anthropic, 2 September 2026).

Last updated: 12 October 2026.

What Anthropic published on 2 September

"Today, we're launching a blueprint to help build commerce agents on Claude", the post begins, "with reference implementations of a shopping agent and a merchant agent for retail, travel, telecom, and ticketing platforms. It also includes a Claude Code plugin to get you started" (Anthropic, 2 September 2026).

Four things went live together: the post; the repository, github.com/anthropics/commerce-agents, Python under Apache 2.0, created 1 September UTC, whose README calls it "a reference implementation; it is not maintained and does not accept contributions"; an engineering deep dive by Ali Shazal and Matthew Koen; and a solutions page with "8 live demos" and the pitch "Win the sprint to holiday season this year" (claude.com, read 10 October 2026).

Open-source reference code, then, not a product. How a brand gets products in front of Claude at all is covered in Claude commerce: how Claude shops and how brands get in; the surface is on Claude for brands.

Two agents, four verticals, one plugin

Quotes are from the post and the repository (Anthropic, 2 September 2026). The shopping agent carries "guardrails designed to constrain prices and products to actual catalog data, and avoids manipulative upsell patterns".

The verticals are fictional: "the only company is ACME". Retail, travel, telecom and entertainment (the post says "ticketing", the repo says entertainment/), with one consumer and one merchant demo each. That is the arithmetic behind "8 live demos": two agents, four verticals, eight web apps, not eight agents.

The plugin, commerce-builder, has "Six skills and four commands" (/scaffold-commerce-agent, /add-commerce-flow, /author-commerce-evals, /review-commerce-agent) and "runs no code of its own" (plugins/commerce-builder/README.md). If you came here wanting Claude Code to build you a store: it scaffolds an agent that talks to a catalog you already have, not a storefront.

Three ways to run it

The README names three runtime paths over the same prompt, skills and tool contracts: the Messages API ("The reference loop"), the Claude Agent SDK ("the SDK running the loop") and Claude Managed Agents ("calling your MCP server"). On the first two the agent calls Python methods on a StorefrontBackend the merchant implements: "11 required methods, plus get_account_context, get_disclosure, and checkout_handoff". MCP enters only on the hosted path; of connectors the README says "None ship". It deploys to "the Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI" (Anthropic, 2 September 2026). What a merchant implements versus what ships is in Claude commerce agents and MCP for merchants; our own server is at connect Tru Commerce from Claude.

Who is named

The post names "Enterprise customers like Shopify, Priceline, and others" and "solutions and ecosystem partners such as Accenture, Mastercard, and Visa". Eleven logos: Accenture, Fetch, Intuit, Klaviyo, Mastercard, Priceline, Shopify, Square, Visa, Wix and Zomato (Anthropic, 2 September 2026).

Shopify did more than lend a logo. Shopify/claude-for-commerce-examples, created 1 September and announced 2 September, is "a storefront shopping agent over UCP and Sign in with Shop, and a merchant agent over the Admin API". Point it at a store and "the agent searches its live catalog, builds a real cart, answers from its policies and FAQs, and hands the customer to the store's own checkout page". Same boundary as Anthropic's: "Nothing here places an order or takes payment: checkout, shipping, and payment all happen on Shopify's own pages" (Shopify README). Where UCP sits next to ACP and AP2 is in ACP, UCP, AP2 and MCP explained.

Zomato is the only Indian company on the page, and the blueprint says nothing about India, UPI or availability; the context is the NPCI and Anthropic UPI shopping pilot reported in February, with a Rs 15,000 cap and BigBasket, Zomato and Swiggy Instamart as participants (BusinessToday, 18 February 2026).

The numbers, with their exact status

No retailer, baseline or window is named for the first two. Angela Jiang, head of product for the Claude platform, gave Reuters the narrower version, "about 30-35% for one partner" (Reuters, 2 September 2026): print the two together or not at all. Adobe's 60% is a conversion rate for all AI-referred retail traffic, same digit, different metric, nothing to do with Claude.

The speed figures are quotes too: "running in days" is Anthropic's; "within fifteen minutes" is Dror Zalika's, Head of Commerce at Wix; "well under an hour" is Ashley Nader's, a Staff Product Manager quoted on the post (Anthropic, 2 September 2026).

Timeline

What did not ship

• A shopping surface inside Claude. "The shopping agent lives inside your app or website" (Anthropic, 2 September 2026); Claude.ai users do not shop through it. Claude in Chrome is a separate consumer feature that "reads the page you're signed in to" and "stops before taking sensitive actions" (claude.com/chrome, read 10 October 2026); the difference is drawn in how Claude picks products.
• A product feed, crawler or schema requirement. The agent reads what the merchant's backend returns, as compact rows capped at 12,000 characters per product family (docs/backends.md). Gupta's reading: "the product page itself does not really come into it" (Scot Wingo on LinkedIn, 11 September 2026).
• Payments. "Nothing in the repo handles a payment credential" (docs/safety.md); StorefrontBackend has no charge method. Reuters: "The tools do not autonomously complete purchases on behalf of customers or merchants" (2 September 2026). The checkout it hands off to is still yours to make finishable, which is what Enable agentic checkout does.
• Ranking. Search is a backend method the merchant implements; the deep dive says agent tools "should call" the production search, ranking, cart, inventory and fulfillment systems, and "the tool boundary is where their logic ends" (Anthropic, 2 September 2026). Any claim about how Claude ranks products is the claimant's, not Anthropic's.
• ACP, UCP or AP2 in Anthropic's code. None appears in the repository; UCP enters only through Shopify's quote and Shopify's repo.
• A merchant directory or listing programme. The Shopify connector in the Claude connectors directory, added April 2026, is a merchant admin tool with 25 tools for products, orders and inventory, unrelated to the blueprint (claude.com/marketplace/connectors/shopify, read 10 October 2026). The storefront-inside-Claude that some vendor guides described in May 2026 appears on no Anthropic page.
• Pricing. Nothing beyond API usage. The post routes to "Contact our sales team"; no fee, revenue share or placement programme exists in the primary pages.

Where this leaves Claude among the assistants a brand has to show up in is in six AI agents your brand needs to show up in; how each of them buys is in AI shopping agents: how they buy.

Where Tru Commerce fits

MCP App Builder deploys a brand's catalog as a live MCP server. Ours connects from Claude on the web and desktop as a custom connector and from Claude Code with claude mcp add; the steps are at connect Tru Commerce from Claude. We are built with Claude Code. We have run this on Gray Porter and a group of US DTC brands: we made the product pages and checkout readable and finishable by an agent, then placed a real Muse order on the store. Agent orders also have a habit of landing in GA4 as Direct: on MyMuse, GA4 reported ₹11.5K/mo of AI revenue against ₹81.2K/mo measured. To see what an agent gets back from your catalog today, book a demo.