Google AI Releases Universal Commerce Protocol (UCP): An Open Source Standard Designed to Power the Next Generation of Agentic Commerce

Can AI retail agents go beyond sending product links and actually complete trusted purchases within a conversation? The Universal Commerce Protocol, or UCP, is Google’s new open standard for agent commerce. It provides AI agents and merchant systems with a shared language so that a purchase inquiry can go from product discovery to order confirmation without custom integration for every merchant and every location.

What problem does UCP solve?

Today, most AI shopping experiences stop at recommendations. The agent integrates links, manages stock checks, coupon codes, and checkout flows from different sites. Google’s engineering team describes this as the N by N integration bottleneck. Each new chat area requires a different function for every merchant and payment provider.

UCP collapses that into a single abbreviation. Platforms such as Gemini or AI Mode search include and protocol. Businesses disclose their commercial conduct once after the UCP. Payment Service Providers and Guarantee Providers include the payment layer. The same protocol can support multiple modes such as shopping, travel, or services.

Key roles and building blocks

The text of the core concepts of the UCP defines four main actors:

• Platform, which is the agent or application that organizes the user journey. Examples include AI shopping assistants and search engines.
• An entity, which is a retailer or service provider and a Retailer of Record.
• Authentication Provider, which handles payment instruments and personal data such as addresses.
• Payment Service Provider, which processes authorizations, captures, and payments.

In addition to these roles, the UCP presents three basic structures:

• Capabilities, such as Checkout, Identity Linking, and Ordering.
• Extensions, such as discounts, fulfillments, or AP2 instructions, that extend the ability by extends field.
• Services, which include transport capabilities such as REST API, Model Context Protocol, and Agent2Agent.

The GitHub repo lists four key capabilities to purchase:

• Checkout handles checkout times, cart contents, total, and tax.
• Identity Linking uses OAuth 2.0 so agents can represent users.
• Order releases lifecycle events for shipments, returns, and refunds.
• Payment Token Exchange connects token and authentication between payment service providers and authentication providers.

The commercial life cycle of an AI agent

Google’s reference and sample repository uses a simple store to demonstrate UCP flow.

A typical agent output looks like this:

1. The agent downloads the business profile from /.well-known/ucpyou get that dev.ucp.shopping.checkout and corresponding extensions are available, and solve schemes for those skills.
2. When a user has linked an account, the agent performs Identity Linking with OAuth 2.0 scopes that allow checkout and ordering learning activities for that vendor.
3. The agent invokes the Checkout capabilities using REST or MCP bindings, pass-through line objects, the buyer’s environment, or any context required. The server returns a payment object with line items, values, and candidate fill options.
4. The agent uses discounts or loyalty benefits by requesting extensions that change the combined checkout schema, and then asks the user to confirm the final order.
5. The payment is routed to a payment processor that understands the schema of a particular payment instrument, such as a tokenized card. If the Payment Service Provider approves the transaction, the business creates an order.
6. The order capability emits webhook events for shipping and purchase orders being shipped, which the agent can present in the chat.

From the user’s perspective, this keeps the entire flow in one conversation with clear consent steps.

Transportation, payments, and security

The specification defines a transport layer with REST bindings, Model Context Protocol, Agent2Agent, and Embedded Protocol that enables a deep merchant payment experience while using UCP data structures.

For payments, UCP includes the Agent Payments Protocol. The payment structure separates the payment instruments from the payment handlers, and executes commands that are hashed to a specific output hash. This design supports binding proof and reduces the risk of token replay, which is important when agents make payments without the user directly in the browser.

Authentication providers issue tokens and hold sensitive card data or addresses. Payment Service Providers consume those tokens and communicate with card networks. UCP keeps these roles clear and uses authentication letters and signatures so that both agents and businesses have confidential evidence of what is authorized.

Key Takeaways

• UCP is an open standard and open source specification from Google that defines a common commercial language for AI agents, businesses, payment providers, and data providers throughout the entire purchase journey.
• The protocol was developed in collaboration with partners such as Shopify, Etsy, Wayfair, Target, and Walmart, and is already accepted by more than 20 ecosystem players including Visa, Mastercard, Stripe, PayPal, Best Buy, The Home Depot, Macy’s, and Zalando.
• UCP models trade-in available capabilities such as Checkout, Identity Linking, and Ordering, as well as extensible modules for discounts, fulfillment, and subscriptions, with vendors and agents negotiating shared capabilities dynamically set by profiles hosted in .well-known/ucp.
• The protocol is transport-agnostic, supporting REST, JSON RPC, Model Context Protocol, and Agent2Agent so that the same power schemes can be reused across back-end services, MCP tool calls to LLM agents, and agent to agent networks without rewriting business logic.
• Payments in UCP includes the Agent Payments Protocol and a modular payment handler design, which isolates payment instruments from administrators and uses transparent instructions, so that agents can withdraw money automatically while maintaining clear evidence of user consent.

Check it out GitHub Repo again Technical details. Also, feel free to follow us Twitter and don’t forget to join our 100k+ ML SubReddit and Subscribe to Our newspaper. Wait! are you on telegram? now you can join us on telegram too.

Check out our latest issue of ai2025.deva 2025-centric analytics platform that transforms model implementations, benchmarks, and ecosystem activity into structured datasets that you can sort, compare, and export.

Asif Razzaq is the CEO of Marktechpost Media Inc. As a visionary entrepreneur and engineer, Asif is committed to harnessing the power of Artificial Intelligence for the benefit of society. His latest endeavor is the launch of Artificial Intelligence Media Platform, Marktechpost, which stands out for its extensive coverage of machine learning and deep learning stories that sound technically sound and easily understood by a wide audience. The platform boasts of more than 2 million monthly views, which shows its popularity among viewers.