Agentic Commerce via MCP

Something major happened last week. OpenAI announced support for Anthropic's MCP and now this will become industry standard and power agentic commerce. I made a demo showcasing how powerful this will be for brands.

A new standard called Model Context Protocol (MCP) is emerging as the crucial link between brands and AI assistants. It's like a a USB-C port for AI applications. I've been testing it and I'm convinced it completely transforms how customers research and buy products.

The Problem With AI Shopping Today

New research shows 63% of websites have received AI traffic, mostly from ChatGPT/Claude/Perplexity. So people are using these AI assistants for research.

But if you've ever tried asking Claude or ChatGPT about products, you've probably seen responses like:

This is painful but it isn't Claude's fault. The AI shopping experience today is fundamentally broken because:

1. AI assistants can only guess based on old training data

2. They can't check current inventory or pricing

3. They force customers to leave the conversation to complete a purchase

4. They can't make genuinely personalized recommendations

MCP Changes Everything

When a brand connects directly to AI assistants through MCP, the customer experience becomes something entirely different:

The difference is night and day. MCP enables:

1. Real-time product information pulled directly from the brand's database

2. Complete details about specific models, features, and availability

3. Contextual recommendations based on the customer's actual needs

4. The ability to purchase products without ever leaving the conversation (though this is going to need checkout support from Shopify)

How MCP Actually Works

Without getting too technical, MCP creates a standardized way for AI assistants to talk directly to external systems. In my testing, I connected Claude to two different MCP servers:

1. A custom MCP server I created for Finisterre for this demo that uses their vectorized product data in Pinecone vector database.

2. A Stripe payment system that lets customers complete purchases (already publicly available MCP server)

Here's what the process looks like:

1. The customer asks about products

2. Claude pings the brand's product database for current information

3. When the customer decides to buy, Claude generates a payment link through Stripe

4. The customer completes their purchase without ever leaving the conversation

RAG: The Engine Powering MCP Product Intelligence

While MCP provides the protocol for connecting AI assistants to brand data, Retrieval-Augmented Generation (RAG) is the underlying technology that makes these interactions truly powerful. Here's how RAG transforms the AI shopping experience through MCP:

How RAG Powers MCP Product Intelligence

At its core, RAG enables Claude and other AI assistants to:

1. Access real-time product information: When a customer asks about a product, the AI doesn't rely on potentially outdated training data. Instead, it retrieves the most current information directly from the brand's vector database through MCP.

2. Understand nuanced product queries: RAG helps the AI interpret complex customer needs by matching their query intent with the most relevant product attributes in the vector database.

3. Make contextual recommendations: By combining the customer's stated preferences with semantically similar product features, RAG powers recommendations that go beyond simple keyword matching.

The Technical Implementation

Behind the scenes, MCP-enabled product discovery works like this:

1. Vector embedding: Product descriptions, specifications, and features are converted into high-dimensional vector embeddings that capture semantic meaning.

2. Similarity matching: When a customer asks a question, their query is also converted to a vector and compared against the product database to find the closest matches.

3. Contextual retrieval: The most relevant product information is retrieved and provided to the AI assistant to incorporate into its response.

4. Augmented generation: The AI generates a natural language response that seamlessly integrates this retrieved information with conversational guidance.

Real-World Example

Consider this interaction:

The difference is dramatic. In the second response, Claude isn't guessing based on general knowledge—it's retrieving specific, current product information that precisely matches the customer's requirements.

Why RAG Makes MCP Exponentially More Valuable

RAG enhances MCP in several key ways:

• Semantic understanding: RAG understands concepts, not just keywords, allowing for natural language shopping

• Product-to-need matching: It can connect customer problems with product solutions, even when they don't use the exact terminology

• Personalization: The system gets better at matching customer preferences to products over time

• Contextual memory: It can maintain the thread of a conversation, understanding how follow-up questions relate to previous inquiries

For brands implementing MCP, integrating a well-designed RAG system means customers can have genuinely helpful shopping conversations rather than merely accessing a database. The AI becomes a knowledgeable sales associate rather than just a search interface.

Why Brands Should Care

If you run an e-commerce brand, here's why MCP should be on your radar:

1. You Control Your Product Information

Without MCP, AI assistants are guessing about your products based on whatever they were trained on. With MCP, you ensure customers get accurate, up-to-date information directly from your database.

2. Genuinely Helpful Product Recommendations

MCP enables AI to understand your customers' specific needs and make tailored recommendations from your product line:

This level of personalization was previously impossible at scale.

3. Dramatically Reduced Purchase Friction

Think about the normal customer journey:

1. Get generic info from AI

2. Visit your website

3. Find the products mentioned

4. Add to cart

5. Go through checkout

6. Come back to AI with more questions

With MCP, it's just:

1. Ask about products

2. Get specific, accurate information

3. Add to cart within the conversation

4. Complete purchase via checkout link (or ideally check out directly within chat)

5. Continue the conversation for post-purchase support

Every reduction in friction means more completed purchases.

4. Better Product Discovery

Without MCP, customers only find what they explicitly ask for. With MCP, contextual discovery introduces them to products they might not have considered:

The AI can guide customers to complementary products that enhance their overall experience.

Shopify's Huge Opportunity

While my demo used Stripe for payment processing, there's an even bigger opportunity for Shopify. They could transform agentic commerce by standardizing MCP checkout across their platform.

Shopify is perfectly positioned for this because:

1. They already power over 2 million online stores, with Shop Pay providing a standardized checkout

2. Their platform already handles payments, inventory, and order management

3. Most small-to-medium brands already trust Shopify with their commerce operations

A Shopify MCP implementation would focus on what they do best (payments and checkout), while allowing brands to develop specialized MCP systems for product recommendations and customer experiences.

The Missing Piece: MCP Discovery

For this to work at scale, AI assistants need a way to discover which brands offer MCP capabilities. Enter the emerging agents.json standard—a machine-readable file that lives at a predictable location:

This file tells AI assistants:

1. What capabilities the brand offers

2. Where those capabilities can be accessed (url of the MCP server to use)

3. What data formats to use

This discovery mechanism would transform AI shopping from brand-specific queries to need-based queries. Instead of asking "Does Finisterre have a 5/4 wetsuit?", customers could simply ask:

The AI could then:

1. Identify this as a shopping intent for cold-water surf gear

2. Discover (and perhaps prioritize) relevant brands with MCP support

3. Query their product databases with the specific parameters

4. Return personalized recommendations from the best-suited brands

Getting Started With MCP

Creating your own MCP commerce server is the first step to agentic commerce:

1. Convert your product catalog into a vector database (like Pinecone)

2. Implement the MCP protocol (Python or TypeScript SDKs are available)

3. Connect with Stripe or other payment providers

4. Deploy for use with Claude, ChatGPT, or other MCP-compatible AI assistants

The First-Mover Advantage

The difference between standard and MCP-enabled AI shopping is so dramatic that early adopters will gain significant advantages:

1. Better customer experiences leading to higher satisfaction

2. Increased conversion rates from reduced friction

3. Clear differentiation from competitors still using traditional channels

4. Rich insights into customer questions and preferences

The brands that implement MCP now will establish themselves as the default options for AI-driven product recommendations. This creates a new form of digital shelf space potentially more valuable than traditional SEO or paid advertising.

Ready to set up MCP for your brand? a.gentic can help. DM Bora on LinkedIn to jam on getting started.