From UX to AX: MCP Servers for AI Shopping Agents

The Shift from User Experience (UX) to Agent Experience (AX)

The State of E-commerce Today

• 51% of Gen Z customers start product searches on AI platforms, not retailer websites
• Estimated $500 billion in annual e-commerce sales through AI agents by 2030
• Gartner predicts 25-30% of all e-commerce will flow through AI agents in the near future
• Open AI reported 340% year-over-year growth in product searches on their platform
• Traditional channels like Google product searches are declining by 23% year-over-year, with 30% drop in click-through rates

The Shift from UX to AX

• Over the last two decades, e-commerce has focused on perfecting user experience (UX)
• But AI agents don't care about visual design, navigation, or page load times
• The fundamental shift is from optimizing for human users to optimizing for AI agents
• Key requirements for agent experience (AX):
  • Good API design
  • Structured, up-to-date, and reliable data
  • Protocols to enable agents to interact with e-commerce data and functionality

Protocols for Agent-Powered E-commerce

Popular Agent Interaction Protocols

• ACP (OpenAI and Stripe): Commerce-specific protocol for product searches and checkout
• AP2 (Google): Protocol for Google payment flows
• A2A (Google): Generic protocol for agent-to-agent communication
• MCP (Anthropic): "Swiss knife" protocol for broader e-commerce functionality

Challenges with Implementing MCP

• Local MCP vs. Remote MCP: Local MCP is for personal automation, not agent commerce
• Security: Handling authentication, authorization, and observability for agent interactions
• Scalability: Agents can perform thousands of requests per minute, unlike human users
• Evolution: Protocols and e-commerce capabilities will continue to evolve rapidly

AWS Bedrock Agent Core Gateway

Simplifying MCP Implementation

• Automatically converts existing APIs and Lambda functions into MCP-compatible tools
• Provides a single, secure endpoint for agents to interact with e-commerce capabilities
• Offers context-aware discovery, only exposing the relevant tools and permissions for each agent interaction

Key Benefits

• Eliminates the need to build MCP servers from scratch
• Handles security, authentication, authorization, and observability out of the box
• Ensures compatibility with evolving protocols and e-commerce capabilities

Best Practices for Implementing Agent-Powered E-commerce

Design Agent-First APIs

• Ensure data is structured, semantically rich, and consistently formatted
• Avoid ambiguity that human users can interpret but agents cannot

Progressive Disclosure

• Start by exposing a few key APIs (e.g., product metadata, pricing, checkout)
• Measure performance and learn from agent interactions before expanding

Multi-Protocol Support

• Prepare for the evolving landscape of agent interaction protocols
• Abstract complexity through solutions like AWS Bedrock Agent Core Gateway

Key Takeaways

• The shift from user experience (UX) to agent experience (AX) is inevitable and happening now
• MCP provides a foundation for enabling agent interactions, but comes with challenges
• AWS Bedrock Agent Core Gateway simplifies MCP implementation and addresses key concerns
• Best practices include designing agent-first APIs, progressive disclosure, and multi-protocol support