Agentic Commerce Protocols: A Technical Guide to UCP, ACP, MCP, and A2A

When customers ask AI assistants for product recommendations, they expect accurate prices, real availability, and a path to purchase. When your systems speak a different language than each AI platform, you miss those conversations. Agentic commerce relies on interoperable protocols so AI assistants can discover, validate, and transact with your brand reliably.

eLLMo AI supports four complementary protocol families that work together to enable the complete agentic commerce workflow. Each protocol serves a specific purpose in the journey from discovery to transaction.

When brands rely on proprietary APIs for AI integration, they face a critical challenge: each AI platform requires custom integration work, creating maintenance overhead and vendor lock-in. Standards-based protocols solve this problem by providing a common language that works across multiple AI surfaces.

Model and assistant independence: By implementing open protocols, your catalog, pricing, and policies become interpretable by ChatGPT, Google AI, Perplexity, and future AI platforms without additional integration work. Your semantics stay durable as platforms evolve.

Trust and control: When AI assistants quote outdated prices or wrong product details, customers lose trust and you lose sales. Verified catalogs with audit trails keep answers consistent with your brand truth.

Time-to-value without replatforming: eLLMo AI activates protocol distribution in under four hours by translating protocol messages to your existing REST or GraphQL APIs. You keep your current CMS, PIM, OMS, and payment systems while becoming protocol-ready.

Model Context Protocol (MCP) establishes how AI assistants discover what capabilities your systems offer and how to invoke them within proper consent boundaries. Think of MCP as the handshake that precedes any commerce interaction.

When an AI agent connects to your brand through MCP, it receives a manifest describing available tools (like product search or inventory check), resources (like your product catalog), and prompts (like policy-aware response templates). The agent knows exactly what it can do and what information it can access, all governed by user consent.

Key MCP capabilities:

• Tool registration with structured schemas defining inputs and outputs
• Resource discovery with URI-based access to data sources
• Prompt templates that ensure responses honor your business rules
• Lifecycle management for enabling, disabling, or deprecating capabilities
• Consent scopes that limit what actions agents can perform

Here's an example MCP capability manifest showing how tools and resources are exposed:

{
  "tools": [
    {
      "name": "get_product",
      "description": "Fetch normalized product data by ID.",
      "inputSchema": {
        "type": "object",
        "properties": {
          "id": { "type": "string" }
        },
        "required": ["id"]
      }
    },
    {
      "name": "search_catalog",
      "description": "Search catalog by query and filters.",
      "inputSchema": {
        "type": "object",
        "properties": {
          "q": { "type": "string" },
          "filters": { "type": "object" }
        }
      }
    }
  ],
  "resources": [
    {
      "uri": "ellmo://catalog/products",
      "title": "Products",
      "mime": "application/json"
    }
  ],
  "prompts": [
    {
      "name": "answer_with_policies",
      "purpose": "Ensure responses honor return and shipping policies."
    }
  ]
}

Universal Commerce Protocol (UCP) solves the product data fragmentation problem. Different ecommerce platforms represent products differently. Shopify has one schema, WooCommerce another, and custom platforms vary widely. UCP provides a normalized, category-agnostic format that AI agents can reliably parse regardless of your underlying platform.

When an agent requests product information through UCP, it receives a consistent payload structure with identity (product IDs, variants), composition (materials, ingredients), specifications (dimensions, technical details), pricing (amount, currency, promotions), availability (in stock, backorder), and policy details (returns, shipping, memberships).

Critical UCP features:

• Normalized attributes work across verticals (fashion, beauty, food, electronics)
• Identity fields ensure the right product is referenced in conversations
• Policy hooks connect to your actual terms so agents quote accurate conditions
• Incremental updates via webhooks or polling keep data fresh
• Versioned schemas allow evolution without breaking existing integrations

Example UCP product payload for a beauty product:

{
  "id": "prod-001",
  "name": "Hydrating Serum 30ml",
  "description": "Lightweight serum with hyaluronic acid and vitamin C.",
  "attributes": {
    "ingredientList": ["Hyaluronic Acid", "Vitamin C"],
    "skinType": "All",
    "size": "30ml"
  },
  "price": {
    "amount": 79.99,
    "currency": "USD"
  },
  "availability": "in_stock",
  "policy": {
    "returnWindowDays": 30,
    "membershipEligible": true
  },
  "updatedAt": "2026-02-19T12:00:00Z"
}

This structure works identically whether the underlying data comes from Shopify, WooCommerce, or a custom PIM. The agent sees consistent semantics every time.

Agent Commerce Protocol (ACP) handles the transaction lifecycle in conversational commerce. Unlike traditional checkout flows where users click through predefined steps, ACP enables multi-turn negotiation where the agent and user can discuss options, apply promotions, adjust quantities, and confirm terms before finalizing the purchase.

ACP maintains conversational state across these interactions, tracking offers, counteroffers, payment authorization, and fulfillment coordination. When a customer asks an AI assistant to "add the serum to my cart and apply my membership discount," ACP orchestrates the entire flow while enforcing your business rules.

Core ACP capabilities:

• Offer construction with line items, discounts, taxes, and shipping
• Multi-turn negotiation preserving conversational context
• Payment orchestration connecting to your existing payment gateway
• Fulfillment coordination with order management systems
• Promotion and membership rule enforcement during offer assembly
• Audit trails for compliance and dispute resolution

Example ACP offer message with applied discount:

{
  "type": "offer",
  "offerId": "off-789",
  "items": [
    {
      "productId": "prod-001",
      "qty": 1,
      "unitPrice": {
        "amount": 79.99,
        "currency": "USD"
      }
    }
  ],
  "discounts": [
    {
      "code": "WINTER10",
      "amount": {
        "amount": 8.00,
        "currency": "USD"
      }
    }
  ],
  "total": {
    "amount": 71.99,
    "currency": "USD"
  },
  "validUntil": "2026-02-29T23:59:59Z",
  "terms": {
    "shipping": "standard",
    "returns": "30d"
  }
}

The agent can present this offer to the customer, who might accept it, ask for expedited shipping, or request to add another item, all handled through subsequent ACP messages.

Agent-to-Agent Protocol (A2A) becomes essential when a purchase involves multiple vendors or requires coordination between specialized agents. Imagine a customer asking for a gift bundle where the main product comes from your store but the gift wrapping and expedited shipping come from partners. A2A enables secure, consented delegation between agents.

A2A handles identity, authorization, rate limits, and signed messages to ensure cross-agent interactions remain secure and auditable. Each agent operates within defined delegation policies, and the user maintains control over which cross-vendor actions are permitted.

A2A use cases and features:

• Shipping quote delegation when fulfillment partners have better rates
• Bundled product assembly across multiple vendor catalogs
• Subscription coordination when products come from different sources
• Identity and authorization with signed message envelopes
• Delegation policies defining what agents can request from each other
• Idempotent retries and delivery guarantees for reliable coordination

Example A2A context update requesting a shipping quote from a specialized agent:

{
  "from": "agent:concierge-1",
  "to": "agent:shipping-quote",
  "intent": "quote_request",
  "payload": {
    "items": [
      {
        "id": "prod-001",
        "qty": 1,
        "weightGrams": 200
      }
    ],
    "destination": {
      "country": "US",
      "postalCode": "94105"
    },
    "speed": "standard"
  },
  "traceId": "7f3e-4b2a-9c1d",
  "timestamp": "2026-02-19T12:10:00Z"
}

The shipping agent processes this request and returns a quote that the concierge agent can incorporate into the final offer to the customer.

You do not need to implement all four protocols immediately. Start with the capabilities that match your current business needs and expand incrementally.

Start with MCP and UCP when your primary goal is making your catalog discoverable and interpretable by AI assistants. This foundation enables organic appearances in AI-generated answers when users ask product questions.

Add ACP when you want to enable conversational commerce transactions, not just discovery. ACP becomes valuable when you see assistant-driven traffic converting and want to own the checkout experience within the conversational flow.

Enable A2A when your products require coordination with other vendors or specialized fulfillment services. A2A is essential for marketplaces, subscription bundles, and any scenario where multiple parties need to coordinate on a single customer transaction.