Quick Start

Get from zero to your first AI shopping session in under five minutes. No account required, no setup needed.

Step 1: Open the Inspector

Head to ucpplayground.com and navigate to the Inspector tab. This is where you connect to any UCP-compatible store and interact with its tools.

Step 2: Enter a Domain

Type a store domain into the domain input field, for example pier1.com. You can use any domain that supports UCP -- no trailing slashes, no https:// prefix needed.

Step 3: Auto-Discovery

Endpoint discovery automatically fetches the store's /.well-known/ucp manifest. This tells UCP Playground which UCP version the store runs, which transports it serves (MCP, REST, A2A, or embedded checkout), what capabilities it declares, and what payment methods it accepts. If the store doesn't support UCP yet, you get a notice saying so instead of an error.

Step 4: Browse and Call Tools

Once connected, the Inspector shows all available tool calls. For MCP stores, you can invoke them manually. For REST-only stores, you get a search interface and product cards with add-to-cart. Common tools:

• search_catalog -- search for products by keyword (MCP and REST)
• get_product or lookup_catalog -- fetch full details including variants and pricing
• create_checkout -- create a checkout session with selected items

This is useful for debugging tool responses and verifying that a store's UCP implementation is working correctly before running an AI agent. REST-only stores work with both the Inspector search and the Agent benchmark. The WEBMCP button shows a different set of tools: the ones the store's own web page registers (see WebMCP: the store page).

Step 5: Switch to the Agent Tab

Click the Agent tab to run an AI model against the store. The connect card asks for a Model (Claude, GPT, Gemini, and others), a Scope toggle (Specific Store, or Shop All Stores in beta), and the store domain -- then hit Connect. If you don't have a store in mind, use the try-pills row beneath the form; flower-shop.local and hotel.local are tagged as test stores, for shopping and for lodging. Once connected, the endpoint card's Connection picker chooses how the agent reaches the store: UCP over MCP or REST, or, if you are signed in, the store page's WebMCP tools.

Step 6: Chat with the Agent

Once connected, type a shopping prompt into the chat composer, like "Find me a woody fragrance under $100," and send it. The agent autonomously calls tools to search, browse, and attempt a purchase, and you see each tool call, response, and decision in real time. When it finishes, the rail shows how far the agent progressed -- from initial search all the way through checkout -- along with token usage and timing data. For a deeper look at what each metric means, see Understanding Results.

Next Steps

Now that you have run your first session, dive deeper:

• Endpoint Discovery -- learn how manifest discovery works and what each field means
• Agent Runs -- configure prompts, models, the connection, and session parameters
• WebMCP: the store page -- run a task through the store page's own tools and compare it with UCP
• Understanding Results -- interpret outcomes, token usage, and timing data