Universal Commerce Protocol (UCP) is an open-source standard, built by Google with more than twenty commerce partners, that lets AI agents discover a merchant's products, build a cart, and complete checkout inside a conversation. Merchants publish a capability manifest, keep their own business logic, and remain the merchant of record throughout.
The interesting part sits below the headline. UCP is a specification with a discovery mechanism, a payments architecture, and a versioning scheme, which makes it possible to reason about who should adopt it and when, with far less guesswork about the shape of agentic commerce.
Universal Commerce Protocol is a shared language between AI agents and merchant backends
UCP standardises the whole commerce journey, from product discovery through checkout and order management, behind one abstraction layer. Google published it as an open-source standard developed with Shopify, Etsy, Wayfair, Target and Walmart, and endorsed by more than twenty global partners including Adyen, American Express, Mastercard, Stripe, The Home Depot and Visa (Google Developers Blog).
Two design decisions matter more than the partner list. Merchants keep their own business logic and remain the merchant of record, so UCP operates as a transport and discovery layer with no marketplace on top. And the protocol is transport-flexible: the same capability can be exposed over a REST API, over Agent2Agent, or over the Model Context Protocol, which is the same plumbing we walked through in our WebMCP explainer.
The N by N integration problem UCP was built to remove
Every new conversational surface currently demands its own integration. Google's own framing names this directly: businesses face an N by N integration bottleneck, forcing them to build bespoke connections for every surface, which slows the whole ecosystem down (Google Developers Blog). Ten agents times ten merchants is a hundred integrations nobody wants to maintain.
A single shared contract collapses that into one integration per merchant and one per agent. This is the same economic argument that made MCP spread quickly through the tooling layer, and it is why the protocol question is worth engineering attention even for teams with no near-term plan to sell inside an assistant. The adjacent B2B version of this problem is covered in our playbook for selling to AI agents.
Capability discovery in UCP runs through a well-known manifest
Merchants publish a JSON manifest at /.well-known/ucp that lists the services they support, the capabilities under each service, the schema for every capability, and the payment handlers they accept. Agents read that manifest and discover features, endpoints and payment configuration at runtime, with no hard-coded integration on either side (Google Developers Blog).
Anyone who has published a robots.txt or an llms.txt will recognise the pattern, and the same discipline applies: the file is the contract, and a stale or malformed one fails silently. Capabilities carry version strings, extensions build on base capabilities such as discounts extending checkout, and everything is fetched fresh. Teams already thinking about how machines read their site should treat this manifest as part of that same surface, alongside the crawl and architecture work underneath it.
Cart, Catalog and Identity Linking pushed UCP past single-item checkout
The first release handled a clean path from discovery to purchase. Google then added three optional capabilities: a Cart option letting agents add multiple items from one store at once, a Catalog capability letting agents pull real-time variants, inventory and pricing from a retailer's catalog, and Identity Linking so a logged-in shopper receives the same loyalty pricing or free shipping they would get on the retailer's own site (Google).
Identity Linking is the one with strategic weight. Without it, buying through an agent means losing member pricing, which gives shoppers a reason to abandon the conversation and open a browser. With it, the agent becomes a legitimate storefront. Adopters choose which capabilities to support, so the manifest doubles as a statement of how seriously a merchant is taking the channel.
Google's UCP reference implementation runs through Merchant Center
The protocol is vendor-neutral, and Google built the first reference implementation on top of it to power buying directly inside AI Mode in Search and the Gemini app. Participating in that specific implementation has a hard prerequisite: an active Merchant Center account with products eligible for checkout, so Google has the product data it needs to surface inventory inside a conversation (Google Developers Blog).
Payments are handled by separating instruments from handlers, which lets existing processors plug in, and every authorization carries cryptographic proof of user consent. Google has also said Commerce Inc, Salesforce and Stripe will implement UCP on their platforms, with a simplified Merchant Center onboarding flow rolling out (Google). For most merchants that means the practical entry point will be a toggle inside the commerce platform they already run.
Deciding whether agentic commerce readiness earns budget this quarter
Three questions settle it for most teams. Does your category get researched conversationally before purchase, does your catalog carry the structured attributes an agent needs to compare variants, and would you accept an order placed without a human seeing your site. A business answering yes three times has a real reason to move now.
Everyone else gets more from the upstream work: clean product data, server-rendered pages, and a machine-readable description of what you sell. That foundation feeds answer engine optimization and any protocol that follows, including ones Google did not write. If you want a read on where your product data actually stands before committing engineering time, we are happy to take a look.
