The Universal Commerce Protocol (UCP) is Google and Shopify's standard for agent-driven shopping, and it launched with a partner list most standards never assemble. I was curious how much of it is genuinely deployed versus endorsed, so I measured it layer by layer and put it in the open, updated monthly, with every number linked to its source: spck.dev/state-of-ucp.
A few things that surprised me.
A served profile is not an adoption
Of roughly 3,000 live UCP endpoints, about 99% are a single platform auto-provisioning an endpoint, not independent adoption decisions. Counting "verified UCP merchants" as adopters overcounts almost everything real.
The frontier lives in a tiny tail
The advanced parts of the spec, message signatures, identity linking, agent payments, appear almost exclusively in a small self-hosted tail. The 99% on the platform default carry none of them. So the spec's ambition is running years ahead of any real, differentiated deployment.
The payments layer is announced, not live
The AP2 payments cohort is the deepest endorsement list of all, and the least deployed: no card network is live yet for agentic card payments. The only production claim in that world is crypto rails, not cards.
"Conformant" is three different questions
Of the normative requirements, some test a merchant server, some bind the buying agent, and some are process/manual. The failures that will actually hurt users, an agent paying a total that does not reconcile or trusting a response it never verified, live on the agent side, which almost nobody measures today.
Method
It is a best-effort, independent read, and I would rather be corrected than confident. The page links every figure to its source, publishes the raw data alongside, and refreshes monthly. If something looks off, corrections are genuinely welcome.
Full report and data: spck.dev/state-of-ucp.
Top comments (0)