Overview
CommerceAI is a shopping agent for AI agents. It gives your agent one MCP tool, shopping.agent, that searches the US product catalog, compares products it already showed, and answers questions about a product.
You send a natural-language shopping request. CommerceAI returns a written answer, the products it recommends, an optional comparison table, the kind of result (outcome), and the credits the call used. The answer comes twice: as self-contained text for your model, and as structured data for your code.
What you can build
| Use case | Example request |
|---|---|
| Product discovery | "Wireless earbuds under $50 with good battery life" |
| Side-by-side comparison | "Compare these two on battery life and price" (with products from an earlier answer) |
| Product Q&A | "Does this one come with a charger?" (with a product from an earlier answer) |
How it works
- Your agent (Cursor, Claude Desktop, or your own code) connects to the CommerceAI MCP server over Streamable HTTP with your API key.
- The agent calls
shopping.agentwith aquery. - CommerceAI runs the shopping agent against the US catalog and returns
outcome,answer_text, the recommended products inanswer_items, andusage. - To continue the thread, your agent sends back
conversation_id. To refer to products it already showed, it sends theiritem_idvalues inselected_item_ids.
One tool, one switch
shopping.agent is the only tool. The query text decides what happens: search, compare, or answer a question. There is no separate search tool or compare tool, and no mode parameter.
Scope
| Supported | Not supported |
|---|---|
| US catalog, English replies | Markets other than US |
| Search, comparison of products already shown, product Q&A | Carts, orders, coupons and vouchers, store account sign-in |
| Text input | Images and image URLs |
| Links to the merchant's product page | Checkout. The shopper buys on the merchant's page. |
Requests for carts, orders, coupons, store accounts, or other markets get a short refusal and are not charged. Image input is rejected with an error. See Limits & Boundary.
Next
- Quickstart: get an API key and make your first call.
- Agent Capabilities: the full request and response reference.