Research, guides and ecosystem updates
Original research on UCP adoption, the State of Agentic Commerce report, guides for each store platform, and the latest spec changes, written for developers and platform teams building for agentic commerce.
UCP Playground Now Runs Personal Agents with Anthropic's Shopping Skills
UCP Playground can now run open-source personal agents, Goose or Hermes, each on its own machine with Anthropic's shopping skills, reaching the store only over UCP. Set the agent's memory, instructions and skills, and read what changed in its files after every run.
· 12 min read
UCP Location Validator: test whether an agent can find your shops
Stores are already declaring UCP location search. Our new Location Validator calls a store's location search and lookup and checks the answers against 32 rules from the spec. Here's what it tests and what the first runs found.
· 7 min read
WebMCP runs in UCP Playground: watch an agent shop a store's website
A website run sends an AI agent to a store's web page instead of its UCP endpoint. The run page shows the chat, every WebMCP call and the agent's browser side by side.
· 8 min read
UCP Playground 0.24: Runs, Inspect and History in one app
UCP Playground 0.24 brings Runs, Inspect and History into one app, follows the UCP checkout lifecycle in every run, and now works without an account.
· 12 min read
Shop through the store page: WebMCP runs in Playground 0.22.0
Playground 0.22.0 runs an agent through the tools a store page registers (WebMCP), keeps every run as a replay with screenshots, and lists page tools beside UCP tools in the Tool Inspector.
· 5 min read
Book a room over UCP: a hotel test store in Playground 0.17.0
hotel.local is a test store for the draft UCP lodging booking capability. Book a room by hand in the Tool Inspector, or let any model book one end to end in Agent runs.
· 5 min read
What the merchant is holding for you: Playground 0.12.0
Three additions to the Tool Inspector and Agent runs. A held-state ledger that mirrors whatever a merchant issued an id for — cart, checkout, hold or reservation — with the merchant's own totals and expiry. A negotiation readout that says why each declared capability is active, passed through or excluded. And buyer context: the spec's country, currency and language signals, sent on every catalog search. Every session now keeps the snapshot it connected with.
· 8 min read
Connect Any Agent to Every Verified UCP Store: the MCP Server, the ARD Finder and Two Skills Are Live
One snippet and your agent gets check-domain, search-all and add-to-cart in its tool tray — backed by the verified list. New API & MCP docs, an Agentic Resource Discovery service for verified stores, an MCP Registry listing and two skills for the agents doing the building.
· 7 min read
One Design System Across the Whole UCP Playground
Inspector, Agent, session replays, Conformance, the docs, the homepage — every surface of the Playground now shares one visual language, one verdict-and-badge vocabulary, one set of tokens. For a tool people use to measure agentic commerce, that consistency isn't cosmetic.
· 4 min read
Introducing UCP Monitoring: Status, Uptime, Score & Alerts in One Dashboard
UCP monitoring, finally in one place. The new dashboard at app.ucpchecker.com watches your business the way agents see it — daily UCP status and uptime checks, UCP Score trends, break alerts, DNS-verified ownership, private staging drift and a watchlist that benchmarks you against the field. Invite-gated beta, free while it lasts.
· 5 min read
Introducing Wire-Level Tool Inspection: See Every Byte Your Store Sends an Agent
Every tool call in a Playground session — live and in replay — now expands to show exactly what crossed the wire: the request we sent, and the raw response the store returned. Plus a refreshed 14-model bench.
· 3 min read
UCP Message Signatures: How Every Request Proves Who Sent It (RFC 9421)
A profile is just JSON anyone can copy; a `UCP-Agent` header is just a string anyone can set. What makes them mean something is the signature on the request carrying them. This is the engine under the agent profile and order webhooks — RFC 9421 covered components, Content-Digest, key resolution by `kid` — and the one encoding bug that breaks almost every first implementation.
· 8 min read