Test stores that speak UCP and WebMCP · no real money, nothing ships · Read the quickstart →
Test stores that speak UCP and WebMCP

Change one thing in a store. See how AI agents react.

Take your own copy of a realistic store. Change the search ranking, a product or the instructions agents read. Send agents to shop both versions, and see what they bought, from the store's own records.

No real platform, no real money: test cards only, nothing ships.

experiment · catalogue rankingSAMPLE

Boost Garden Hat Box for gift searches

Agents bought itA 8%B 41%+33 points
Stayed within budgetA 97%B 84%−13 points
  1. 4.8 sagentsearch_catalog "teacher thank you gift"
  2. 4.8 sstoreboost rule applied · hat box 4th → 1st
  3. 15.1 sagentcreate_checkout hat box × 1
  4. 15.1 sstore$72.00 against “nothing over $50”

240 agent checkouts from UCP Playground, split between A and B.

How it works
01Copy a storeStart from a reference store: Retail, Stays or Food. Your copy gets its own address and nobody else's tests touch it.
02Change one thingA is your store as it is. B is the same store with a single change, so any difference in what agents do comes from that change.
03Send agents to shop bothA and B each get their own address. Point UCP Playground, a CLI or your own agent at them, and each store records what happens.
Experiments

What you can test

Catalogue rankingDoes moving a product up search change what agents buy?
Agent instructionsRewrite a tool description. Do agents search better?
Product dataA badge, a discount, a better description. Which ones do agents notice?
See the experiment setup →Pick a store and one change; A and B get their own addresses.
Results

What you see after every run

Merchants and platformsTry a change on a store like yours before you ship it to your real one.
Agent buildersTest your agent against stores that always answer, and against ones that break on purpose.
ResearchersRun studies anyone can repeat, on stores no platform owns.