Understanding MCP, ACP and UCP in agentic commerce: A merchant’s guide

Agentic commerce offers brands and retailers an array of opportunities, but it’s not without challenges. One of them is understanding and implementing multiple protocols that fulfill specific roles, such as MCP, ACP and UCP. 

Let’s break them down in a way that makes sense and see why retailers need to abstract all protocols into a single commerce layer, so they can focus on strategy, growth and customer experience.

MCP is a protocol created by Anthropic that standardizes how AI agents securely connect to external data sources, tools and APIs. While it can technically be used by individual AI models, such as ChatGPT, Gemini and Claude, its strongest use case is enabling autonomous agents to dynamically discover, negotiate and interact with unfamiliar systems through a structured process.

This makes MCP particularly powerful in scenarios where an AI agent must independently determine how to authenticate, interpret capabilities and execute actions across heterogeneous environments.

This means AI agents can access live, contextual information beyond their training data while respecting privacy and permission boundaries. In practical terms, MCP acts as a bridge between agent models and the real-world systems they interact with, enabling them to access commerce data, execute tasks and retrieve up-to-date information safely and consistently.

What MCP solves: 

• AI system integration with secure context, tool access and real-time data for agents, e.g., an agent can remember that a shopper prefers eco-friendly brands, check inventory levels or loyalty program eligibility before making recommendations. 

• Autonomous shopping agents buying on behalf of customers, including full checkout flows.

What merchants care about:

• A smart agent increases relevance, improves customer experience and drives more informed purchase decisions.

• AI agents can manage the customer journey end-to-end while your business remains in control as the merchant of record.

ACP, created by OpenAI and Stripe (with support from commercetools), focuses on secure, instant checkout, primarily for single-item purchases within GenAI environments like ChatGPT and Microsoft CoPilot. 

To date, ACP has been designed for single-item purchases within the OpenAI ecosystem, using a delegated payment model for security and ease of use, and integrating with Stripe and PayPal. 

What ACP solves: 

• Optimized for payments within the GenAI channel, using delegated payment models via Stripe.

• Allows customers to complete purchases without leaving the GenAI interface.

• Ideal for ecosystems like ChatGPT or Microsoft-integrated AI experiences.

UCP, developed by Google and partners including Walmart and Target, standardizes the entire shopping lifecycle for AI agents, from discovery and comparison to checkout, order tracking and post-purchase support.

As the name suggests, this protocol provides a “universal” language that standardizes how AI agents interact with merchant systems. Merchants can publish their capabilities and product details once, and any UCP-compliant agent can use them. 

What UCP solves: 

• Publish your products and capabilities once, and any UCP-compliant AI agent can access them.

• UCP-compliant AI agent can manage carts, check out securely and even handle returns or refunds.

• UCP supports multiple Google platforms, including Google Search AI Mode and the Gemini app.

• Designed to work with MCP and other Google-built protocols, including A2A (agent-to-agent protocol) and AP2 (agent payments protocol). 

• Broader focus than ACP, it’s designed for long-term multi-agent interoperability. 

Why merchants care about:

• Your products become discoverable across multiple AI platforms without multiple integrations.

• AI agents can manage the customer journey end-to-end.

• Supports multiple payment providers like Google Pay, PayPal and others.

Whether your business uses only one or multiple protocols for agentic commerce depends on your goals, use cases and commerce workflows. Consider these three scenarios: 

1. Build an AI agent for back-office work, analysis or customer support 

MCP lets agents retrieve information from systems, maintain context across tasks and support analysis without touching transactions. 

2. Handle customer-facing buying actions through AI

MCP gives agents a universal way to manage carts and complete checkout, while you stay in control as the merchant. 

3. Build an AI agent to execute purchases autonomously

You may need all three protocols: MCP gives a more universal agent compatibility, including transaction processing. UCP enables agents implemented in accordance with Google standards, including transaction processing. For the ecosystem centered around ChatGPT/OpenAI and Microsoft, ACP is the protocol to go for.  

For most retailers, it’s simply not feasible to build, maintain and update integrations for every emerging agentic commerce protocol on their own. Each new use case you want to implement and its corresponding standard — UCP, ACP, MCP and others — requires significant engineering effort, ongoing updates and specialized expertise. 

With commercetools, merchants can implement agentic commerce across both GenAI channels and brand-owned experiences using commercetools AI Hub and Agent Gateway, respectively. 

By leveraging commercetools, retailers can stay ahead of emerging protocols without the heavy lift — enabling agentic commerce experiences across any channel.