Agentic Commerce: MCP, A2A, and UCP Explained

The next wave of e-commerce will not be driven by humans browsing websites. It will be driven by AI agents making purchases on behalf of humans. An agent that books your flights, compares insurance quotes, orders office supplies, and negotiates vendor contracts — all autonomously, all using structured protocols to communicate with businesses.

This is agentic commerce. It is not theoretical. The protocol infrastructure is being built right now by Google, Anthropic, Shopify, and the open-source community. Three protocols form the foundation: MCP (Model Context Protocol), A2A (Agent-to-Agent), and UCP (Universal Commerce Protocol). Understanding them now gives small businesses a structural advantage before the market shifts.

I deployed agent-discovery files and commerce-ready configurations across all 52 sites in our network. The process took a weekend. When AI shopping agents start evaluating businesses, our sites will be ready. Most competitors will not be.

The Agentic Commerce Landscape

Today, a consumer who wants to buy something opens a browser, searches for products, visits websites, compares prices, reads reviews, and clicks "buy." This process takes minutes to hours and involves significant manual effort.

In the agentic commerce model, the consumer tells their AI agent what they want. The agent searches for products, evaluates options, compares prices, checks reviews, negotiates if applicable, and either makes the purchase or presents a shortlist for the consumer to approve.

Shopify's president described this as a "fundamental shift in commerce infrastructure" that will reshape how small businesses reach customers. Instead of optimizing for human eyeballs on a webpage, businesses will optimize for AI agent evaluation — structured data, clear pricing, transparent policies, and machine-readable product descriptions.

The three protocols that enable this are complementary layers in a stack:

Protocol 1: MCP (Model Context Protocol)

MCP, developed by Anthropic, is a protocol that standardizes how AI models connect to external data sources and tools. Think of it as a universal adapter that lets an AI agent plug into any compatible service.

In the commerce context, MCP allows an AI agent to:

• Query your product catalog in real time
• Check inventory and availability
• Retrieve pricing including any dynamic discounts
• Access your return policy, shipping rates, and terms of service
• Submit orders through a structured API

For a small business, implementing MCP means creating an MCP server — a lightweight API endpoint that responds to agent queries with structured data about your products and services.

What MCP Looks Like in Practice

An MCP server for a small business might expose these capabilities:

{
  "name": "The $97 Dollar Launch Store",
  "capabilities": [
    {
      "name": "search_products",
      "description": "Search available books and digital products",
      "parameters": { "query": "string", "category": "string" }
    },
    {
      "name": "get_product_details",
      "description": "Get full details including price, format, and availability",
      "parameters": { "product_id": "string" }
    },
    {
      "name": "get_shipping_info",
      "description": "Get shipping options and costs",
      "parameters": { "destination_country": "string" }
    }
  ]
}

When an AI agent encounters this server, it knows exactly what data it can request and how to request it. No scraping. No guessing. No parsing unstructured HTML.

Protocol 2: A2A (Agent-to-Agent)

Google's A2A protocol enables AI agents to communicate with each other. While MCP connects agents to tools and data, A2A connects agents to other agents.

In commerce, this matters because the transaction often involves multiple agents: a buyer's agent negotiating with a seller's agent, a logistics agent coordinating delivery, a payment agent handling the transaction.

A2A uses an "Agent Card" — a JSON file at /.well-known/agent.json that describes the agent's identity, capabilities, and communication preferences. When a buyer's agent wants to interact with a seller's agent, it reads the Agent Card to understand what interactions are possible.

The Agent Card for Commerce

A commerce-ready Agent Card declares capabilities that are relevant to AI shopping:

{
  "name": "The $97 Dollar Launch",
  "description": "Digital products and books for launching businesses under $100",
  "url": "https://the97dollarlaunch.com",
  "capabilities": [
    "product-catalog",
    "digital-delivery",
    "pricing-inquiry",
    "bulk-licensing"
  ],
  "interactionModes": ["structured-query", "natural-language"],
  "paymentMethods": ["stripe", "gumroad"],
  "responseFormat": "application/json"
}

This card tells any AI agent: we sell digital products, here is what we offer, here is how to interact with us, and here is how to pay.

Protocol 3: UCP (Universal Commerce Protocol)

UCP is the newest and most ambitious of the three. It is a proposed open standard specifically designed for AI-to-business commerce transactions. While MCP and A2A are general-purpose protocols that can be applied to commerce, UCP is built from the ground up for buying and selling.

UCP aims to standardize:

• Product discovery — how agents find and evaluate products
• Pricing and availability — real-time pricing with support for dynamic discounts, bundles, and negotiation
• Transaction execution — order placement, payment processing, and confirmation
• Post-purchase — shipping tracking, returns, refunds, and customer support

The protocol is still in early development, but early adopters are already implementing draft specifications. The businesses that have UCP-compatible endpoints when the protocol reaches mainstream adoption will have a first-mover advantage.

Why Small Businesses Should Care Now

The common objection is "AI shopping is years away." Here is why that thinking is dangerous:

AI agents are already browsing the web. ChatGPT, Perplexity, and Claude already search for products and services when users ask them to. They cannot complete transactions yet (in most cases), but they recommend businesses and products based on what they find. Businesses with structured, machine-readable product data are already getting recommended over businesses without it.

The infrastructure is being built by the largest companies in tech. When Google, Anthropic, and Shopify all invest in the same category simultaneously, adoption tends to happen faster than expected.

Implementation is cheap and low-risk. Adding an Agent Card, structured product data, and basic MCP compatibility to your website costs nothing and takes hours, not months. The worst case is that agentic commerce takes longer to arrive than expected and you have a slightly better-organized website. The best case is that you are agent-ready when the wave hits and your competitors are scrambling.

Implementing Agentic Commerce Readiness

Step 1: Deploy an Agent Card

Create a JSON file at /.well-known/agent.json describing your business, products, and capabilities. This is the minimum viable implementation — it costs nothing and takes 15 minutes.

Step 2: Implement Structured Product Data

Add Product schema (Schema.org/Product) to every product page. Include:

• Product name and description
• Price and currency
• Availability (InStock, OutOfStock, PreOrder)
• SKU or identifier
• Image URL
• Brand and manufacturer
• Aggregate rating if available

This structured data is already valuable for Google Shopping and rich results. Making it agent-ready is a natural extension.

Step 3: Create a Machine-Readable Product Feed

Publish a JSON or XML product feed at a known URL (e.g., /products.json). This feed should include all products with complete details. AI agents can ingest this feed to understand your full catalog without crawling individual pages.

Step 4: Add llms.txt

Your llms.txt file should include a section describing your commerce capabilities:

## Commerce
- Products available: books, digital tools, templates
- Pricing: transparent, listed on product pages
- Payment: Stripe, Gumroad
- Digital delivery: immediate download after purchase
- Refund policy: 30-day money-back guarantee

Step 5: Monitor Agent Traffic

Watch your server logs for traffic from AI agents. Look for user agents containing "GPTBot," "Perplexity," "Claude," "Anthropic," or "AI." Track which pages agents visit most frequently — these pages are your de facto agent-facing storefront.

The Competitive Window

Agentic commerce is in the same position that e-commerce was in 1998. The early adopters who built online stores before mass adoption reaped enormous advantages. The businesses that build agent-compatible infrastructure today will be positioned to capture AI-directed commerce before their competitors understand what happened.

The implementation cost is minimal. An Agent Card, structured product data, and a product feed can be deployed in an afternoon. The protocols are open standards — no vendor lock-in, no subscription fees, no proprietary platforms.

For our 52-site network, the entire agentic commerce implementation took one weekend. Every site now has an Agent Card, structured product schema, and a machine-readable product feed. When AI shopping agents arrive in force, our sites will speak their language.

The businesses that wait will be invisible to the agents that matter.

For the complete agentic commerce readiness guide and the full $97 business launch blueprint, see The $97 Dollar Launch and The $100 Dollar Network.