Quick start
Get from nothing to your first agent run in a few minutes. You need an account to start runs; the public Inspector works without one.
Step 1: Sign in and open Runs
Sign in at ucpplayground.com and open Runs in the navigation. Runs is where a run starts, with your recent runs at the store under it.
Step 2: Choose a store
The first time, Runs asks Where should the agent shop? Type a store domain (no https://, no trailing slash) or pick one of the test stores: flowers.ucpsandbox.com (Retail), stays.ucpsandbox.com (Stays) or food.ucpsandbox.com (Food). Test stores work end to end and take a test card.
Runs reads the store's UCP profile on the spot (the file at /.well-known/ucp that tells agents what a store supports) and shows what it found; see Endpoint discovery. If the store has no UCP profile, you can check whether its website offers tools the agent can use instead.
Then choose Who the agent is (UCP platform identity, the default, or Unsigned) and check Who it shops for (country, currency and the other buyer details), and click Continue. The store is remembered, so next time you go straight to the composer. Switch stores at any time from the store in the header.
Step 3: Give the agent a task
Runs asks What should the agent do at your store. Click one of the examples, Pick, Buy or Rely, or type your own task in the message box. The pills under the box set the run:
- Model: the model that plays the agent, grouped by provider
- How the agent reaches the store: UCP (the store's APIs) or Website (the store's website, in a browser)
- Who it shops for: the buyer details searches carry
- Who the agent is: UCP platform identity or Unsigned
The Info pane on the right shows what the run will use: your choices, the store and the agent.
Step 4: Send
Press Send (or Enter). The agent starts calling the store's tools, and the run opens on its own page when it first replies. The chat on the left shows what the buyer sees. The pane on the right holds the detail: Trace (every call, kept whole), Discovery and negotiation, Timeline and Info. When the agent asks the buyer something, reply in the box under the chat.
Step 5: Read how it ended
Every run ends as one of: Searched only, Built a cart, Reached checkout, Tried to place the order, Order placed, or Failed. When a call fails, the run says whose fault it was: the store, the agent, or ours. See Understanding results.
Want to look at a store's tools by hand first? Open Inspect, choose the store and click Start, then send each call yourself. Without an account, use the public Inspector. See Tool inspector.
Next steps
- Endpoint discovery: how the UCP profile is read and what each field means
- Agent runs: tasks, models, how the agent reaches the store, and the run page
- WebMCP: the store page: run a task through the store's website and compare it with UCP
- A2A: the store's agent: send checkout to the store's own agent over A2A
- Embedded: the store's checkout: finish in the store's own checkout page
- Understanding results: endings, tokens, cost and timing
- Leaderboard: how models do across stores
Something wrong on this page?