Agentic Commerce

Every Jumpseller store is reachable by AI agents, with nothing to install. This page is the map: what each surface is for, which one to reach for, and what is already switched on.

If you are building an AI assistant that works on Jumpseller — writing themes, calling the API — you want the AI Toolkit instead. This page is about agents that shop a store.

Which surface do I want?

I want to…	Use
Let a shopper’s AI browser act on the page it is on	WebMCP
Read a store’s catalogue from my own agent or app	Storefront MCP
Integrate through the open agentic-commerce standard	Universal Commerce Protocol, below
Manage a store: products, orders, customers	Admin MCP

The first three are public and need no credentials. The fourth authenticates as you.

1. Discover

Before anything can shop your store, it has to find out what the store offers.

• /llms.txt and /agents.md — a plain-text description of the store and the endpoints an agent can use. Generated from the tools that are actually registered, so it never describes a tool that no longer exists.
• /.well-known/mcp.json — MCP discovery metadata.
• /.well-known/ucp — the store’s Universal Commerce Protocol profile: which capabilities it supports, at which versions, and the endpoints that serve them.

All four are public and served by every store.

2. Browse and search

Storefront MCP — POST https://your-store.com/api/mcp

A public, read-only MCP server on every store. Six tools covering store info, products, search and categories. No authentication. Use it when your agent is not running inside the shopper’s browser: a chatbot, a shopping assistant, a comparison service.

UCP Catalog and Cart — POST https://your-store.com/api/ucp/mcp

The same catalogue through the Universal Commerce Protocol: search_catalog, lookup_catalog and get_product for the Catalog capability, and create_cart, get_cart, update_cart and cancel_cart for the Cart one — all in UCP’s own shapes. Use it if you already speak UCP and want one integration that works across platforms.

Both read the same catalogue, and an id from one is an id on the other. They are separate endpoints because the tools mean different things: UCP’s get_product needs a meta envelope naming the agent, and answers in UCP’s Product shape rather than ours.

3. Shop in the page

WebMCP — no endpoint; the store registers tools in the browser.

When a shopper opens a store in an AI browser, the page hands the assistant eleven tools: search the catalogue, browse categories, show a specific variant, fill the cart, apply a coupon, open checkout. They run in the shopper’s own session, so the assistant works on the cart the person is looking at.

This is the one that changes conversion, because it is the one that runs while somebody is actually shopping.

4. Check out

The shopper does. No tool on any of these surfaces can place an order or enter payment details, and inside checkout no tools are offered at all. An agent that built a UCP cart is handed a continue_url for the person to finish in; an agent in the page opens checkout and steps back. Either way an agent’s job ends at handing the person a checkout page.

5. Manage the store

Admin MCP — https://mcp.jumpseller.com

Read and write access to products, orders, customers, categories and pages, authenticated with an API token or OAuth 2.0. This is the merchant’s own agent, not a shopper’s.

What is on by default

Surface	On by default	Authentication
llms.txt, .well-known/mcp.json, .well-known/ucp	Yes	None
WebMCP	Yes, every store, every theme	None — the shopper’s session
Storefront MCP	Yes	None
UCP Catalog and Cart	Yes	None
Admin MCP	Needs credentials	API token or OAuth 2.0

What we have not built yet

Said plainly, because it is more useful than a roadmap:

• UCP Checkout, where an agent completes the purchase itself. That is a separate decision about who takes the money, and not one we have taken. Cart is built: an agent can fill one and hand the shopper back a link to finish in.
• Order tracking for shoppers’ agents. The Admin MCP covers orders for merchants; there is no equivalent for a customer’s own assistant yet.
• A policies and FAQ tool, so an assistant can answer “what is your returns policy” without reading the page.

Where to next

• WebMCP — the in-page tools, and how to test them
• Storefront MCP — the public endpoint and its six tools
• Admin MCP — managing a store from an agent
• AI Toolkit — building on Jumpseller with an AI coding assistant
• REST API — when you want an ordinary integration instead