TensorPass

Use cases

An AI shopping assistant that runs on the shopper's own AI

How a store answers product questions with each shopper's own AI, which MCP tools it serves, which interface it picks and how checkout stays on its pages.

Published on . Updated on

Traffic from AI assistants to US retail sites is forecast to grow 130% over the 2026 holiday season, and 159% on Thanksgiving. Of the shoppers who used AI to shop online, 77% feel more confident in their purchase and 69% are less likely to return it. AI Connect, the TensorPass SDK a website installs, puts that assistant on the store's own pages. The shopper's model reads the catalog through the store's tools and answers on the shopper's own account through AI Wallet, so the store pays no inference.

Questions shoppers bring to an AI

Amazon's Rufus, its shopping assistant, served more than 300 million customers in 2025 and delivered nearly $12 billion of incremental annualized sales. AI and agents influenced 20% of global Cyber Week orders in 2025, worth $67 billion, and retailers running their own agents grew sales 32% faster. On Shopify, traffic from AI assistants grew 8 times in the first quarter of 2026, and orders from AI searches grew nearly 13 times.

Each question below maps to a tool call on the store's server:

Searches run on Shopify's own catalog data convert at twice the rate of searches run on scraped data. An assistant on the store reads that same catalog data through its tools.

Catalog tools served over MCP

The Model Context Protocol (MCP) is the open standard through which a model calls a site's tools. The AI Connect Shopify app serves three read only tools on the store's own domain, at the app proxy path /apps/aiconnect/mcp, and reads the store through the Storefront API.

ToolWhat it returnsQuestion it serves
search_productsThe matching products, each with its handle, title, price range, stock and page address"Running shoes under $120"
get_productOne product in full: description, options, each variant with its price and stock, collections"Is the blue one in stock in size 9?"
list_collectionsThe store's collections, each with a summary and its page address"Gifts for a ten year old"

A store on its own stack writes the tools its catalog needs. A list_categories tool returns the categories, sizes and price bands the catalog accepts, and the model then fills each filter with a value the catalog holds. The MCP guide sets the design rules: a verb and a noun for each name, every enum value explained, every input bounded, and the page address returned with each result.

Which interface suits a store?

The SDK ships five ready made interfaces, described under interface modules. Four of them suit a store:

Checkout on the store's own pages

The Shopify app's tools read the catalog, and the cart and the payment run in the store's own checkout. Each product in the answer links to the page address its tool returned. The answer guard of the SDK lets a link through when it points to the page's own origin, to a host the store lists, or to an address a tool returned as a field. A shopper who clicks lands on the store's product page.

In March 2026 OpenAI scaled back Instant Checkout, and buyers who find a product in ChatGPT now check out on the merchant's own store. Shopify moved its catalog and cart tools to /api/ucp/mcp, and its Storefront Catalog MCP there expects an agent profile in the arguments of every call. The shopper's model has no way to add that profile, and the AI Connect app serves its own three tools for that reason.

A store whose tools return product reviews passes third party text to the model. The SDK closes every tool result in an envelope sealed with a random salt drawn for each conversation, and instructions planted in a review stay inside it as data. This is the guard against prompt injection. The app's admin page holds the store's guards for the catalog: up to 50 patterns of names and content the merchant keeps out of the chat.

Next steps for a store

On 16 November 2026 the SDK packages of version 0.1.0 are on npm, and AI Wallet waits on the Chrome Web Store's review while a visitor can install it by hand from https://tensorpass.app/wallet. A shopper without AI Wallet gets the assistant through a sign in with OpenRouter, or with the TensorPass plan on a site that passes a site key.

Sources

Questions

Who pays for the answers of the shopping assistant?
Each answer runs on the shopper's own ChatGPT, Claude or Gemini account through AI Wallet and counts against that account's plan or credits. The store pays for its MCP server and its hosting.
Does the assistant place orders or take payments?
The three tools of the AI Connect Shopify app read the catalog and change nothing. The answer links each product to its page on the store, and the shopper adds it to the cart and pays in the store's own checkout.
What does a shopper need to use the assistant?
Chrome with the AI Wallet extension and one AI account connected in it. On 12 November 2026 the wallet loads unpacked and its Chrome Web Store item is a draft that has not been submitted.
Does the shopper's AI provider see the whole catalog?
The provider receives the conversation, which holds the tool results of that conversation and nothing else. The catalog stays on the store's server, and the model reads only what search_products, get_product and list_collections return.
Have it done in 24 hoursRead the quickstart