A practical, technical example of using MCP / a2a and AI in eCommerce

This week, we dig into how eCommerce agents collaborate through MCP to drive smarter cross-sells and upsells. But before we get tactical…. where do these agents actually come from?

You’ve got three paths:

1. Build your own in house.

2. Buy vertical specific agents from SaaS vendors.

3. Tap into a growing wave of open source agents and make them yours.

Open source will be a sleeping giant in this space.

Let’s break it all down:

Dynamically present smart cross-sells and upsells during and after purchase to increase AOV and repeat rate.

1. What the Function Does

This function identifies products or bundles to suggest... either in cart, post purchase, or via follow up, to drive incremental revenue.

It should:

• Predict which products pair well

• Decide whether to upsell (higher-end version) or cross-sell (complementary)

• Inject offers into the shopping and post-purchase flow

• Learn from results and optimize combinations

2. Agents Involved

3. How MCP and a2a Work Together

MCP (Model Context Protocol):

• Stores session context: cart contents, purchase history, affinities, past upsell performance, inventory levels

• Ensures all agents work off a single source of evolving truth

a2a (Agent-to-Agent):

• Enables fast, event-based handoffs: “Cart updated → get recommendation → generate offer → inject offer into checkout”

4. Step-by-Step Walkthrough

Step 1: Customer Adds Product to Cart

• Commerce Agent detects product added

• It updates MCP with cart contents and session ID

Step 2: Trigger Cross-Sell/Up-Sell Evaluation

• Commerce Agent → a2a → Context Agent "get_context(customer_id, cart)"

Context Agent checks:

• Past behavior (e.g. category affinities, price sensitivity)

• Purchase frequency

• Open-to-buy logic

Writes session context into MCP:

{

 "customer_id": 123,

 "cart": ["running shoes"],

 "context": {

 "preferred_price_band": "$80-$120",

 "affinity": ["running accessories"],

 "recent purchase": "none"

 }

}

Step 3: Generate Recommendations

• Commerce Agent → a2a → Product Agent "recommend_cross_upsell(cart, context)"

Product Agent uses:

• Collaborative filtering

• Historical bundles

• Best margin or overstock logic

Suggests:

• Cross-sell: Moisture-wicking socks

• Upsell: Premium version of shoe with better support

Writes both to MCP with metadata (score, margin, availability)

Step 4: Dynamic Offer Calculation

• Commerce Agent → a2a → Offer Agent "generate_offer(product_id, context)"

Offer Agent calculates:

• Time-sensitive discount (e.g. $5 off socks if bought now)

• Bundle price (e.g. 10% off if upgraded shoe + socks)

Writes offers into MCP

Step 5: Inject into UI

• Commerce Agent → a2a → UI Agent "inject_offer(cart_ui, offers)"

• UI Agent personalizes layout and tone

• Places inline cross-sell (e.g. “Add socks and save $5”) and upsell module (“Upgrade to X for just $15 more”)

• Writes module details and position into MCP

Step 6: Conversion + Feedback

• Commerce Agent and Checkout Agent track outcomes (accepted, skipped, time spent)

Analytics Agent records:

• Uptake rate

• AOV lift

• Profit margin

• Drop-off points

Analytics Agent → a2a → Retention Agent "log_success_and_use_in_future_lifecycle(customer_id, cross_sell_type)"

Step 7: Follow-up

If offer was skipped or declined:

• Retention Agent may trigger a post-purchase follow-up ("Still thinking about those socks? Get them with free shipping.")

MCP is updated with session learnings:

• “Customer ignored upsell but responded to cross-sell when discounted 10%”

5. Part of the Equation

This function is represented as:

cross_upsell_function = f(cart, customer_context, recommendation, offer, delivery, feedback)

Each component is handled by agents:

• cart = Commerce Agent

• customer_context = Context Agent

• recommendation = Product Agent

• offer = Offer Agent

• delivery = UI Agent

• feedback = Analytics + Retention Agents

MCP = shared source of context, history, and outcomes a2a = real-time coordination across agents for smart decision-making

End Result:

You get a fully autonomous, real-time cross-sell and upsell system.... smartly deciding what to show, when, and how, all based on a single source of truth and self-optimizing over time.

1. Google Unveils Agent-to-Agent (A2A) Protocol to Enhance AI Interoperability in Ecommerce

At Google Cloud Next 2025, Google introduced the Agent-to-Agent (A2A) protocol, an open standard designed to enable seamless communication between AI agents across different platforms and vendors. This advancement allows for more cohesive automation in ecommerce operations, such as inventory management and customer service, by facilitating collaboration between diverse AI systems.​

🔗 Read more on Google's Developer Blog

2. Amazon's AI-Powered Shopping Assistant 'Rufus' Enhances Customer Experience

Amazon is developing advanced AI technologies to enhance its ecommerce platform, aiming to create autonomous AI shopping agents that assist customers by recommending products, adding items to their cart, or even making purchases on their behalf. The company's existing AI, Rufus, already provides comprehensive shopping guides and handles customer inquiries using a proprietary large language model. Future developments could include AI agents performing complex tasks beyond shopping, such as managing daily chores.​

🔗 Read more on WIRED

4. AI's Transformative Role in Ecommerce: What to Expect in 2025

Artificial intelligence is fundamentally reshaping the ecommerce landscape—enhancing customer experiences, streamlining operations, and addressing emerging challenges. From hyperpersonalization and conversational agents to visual search and smarter inventory management, these advancements will define the next era of online shopping.​

🔗 Read more on Digital Commerce 360

5. Emerging Synergies Between Large Language Models and Machine Learning in Ecommerce Recommendations

Recent research explores how large language models (LLMs) can be integrated with machine learning techniques to enhance ecommerce recommendation systems. By leveraging LLMs' capabilities in understanding and generating language, ecommerce platforms can provide more accurate and personalized product recommendations, improving customer satisfaction and sales.​

🔗 Read the full study on arXiv

6. AI Enhances Personalized Shopping Experiences Across Retail Platforms

Retailers are increasingly adopting AI to optimize operations, personalize marketing, and improve customer experience. For instance, Victoria’s Secret has seen a significant boost in marketing metrics by using AI for personalized emails, and Swarovski has successfully integrated AI in customer service and search functionalities, resulting in improved sales and customer satisfaction. Similarly, Saks Global and Caleres have enhanced customer service and shopping experiences through AI-powered tools.​

🔗 Read more on Vogue Business