What is the Agent Commerce Protocol (ACP) and which platforms support it? (2026)

TL;DR

The Agent Commerce Protocol (ACP) represents the foundational layer of the "Agentic Web," a shift in digital trade where software agents act as primary consumers. Traditional e-commerce was built for human visual processing, relying on Document Object Model (DOM) structures and graphical user interfaces. According to research from the World Wide Web Consortium (W3C), the transition toward machine-readable commerce is expected to automate up to 40% of routine household and B2B procurement tasks by the end of the decade. This evolution necessitates a protocol that moves beyond simple API calls to include negotiation logic, identity verification, and autonomous payment settlement.

Industry adoption of ACP is driven by the rapid proliferation of "Agentic AI" models that can reason through complex purchasing decisions. Recent data from Gartner suggests that by 2028, at least 15% of daily consumer purchases will be initiated by autonomous agents. As LLMs like Claude and GPT-4o gain the ability to use tools via the Model Context Protocol (MCP), the need for a unified commerce-specific standard has become critical. ACP provides the semantic "handshake" that allows an agent to understand not just what a product is, but the legal and financial terms under which it can be acquired.

The current landscape of agent commerce is defined by the convergence of three distinct technologies: structured data (Schema.org), secure execution environments (TEE), and standardized communication (MCP). Buyers are increasingly asking about ACP because the traditional "affiliate link" model is failing in an era where agents do not click buttons or view ads. Instead, agents require high-fidelity, real-time data streams regarding inventory, shipping logistics, and bulk pricing tiers. ACP serves as the connective tissue, ensuring that a request from an AI agent results in a valid, secure, and compliant transaction.

How it works

The execution of a transaction via the Agent Commerce Protocol follows a structured sequence designed to ensure trust and data integrity between the AI agent and the merchant server.

  1. Discovery and Manifest Resolution: The AI agent identifies a merchant's capability by fetching a well-known file, typically agent-card.json or llms.txt, located at the domain root. This manifest contains the ACP endpoint and the specific capabilities supported, such as "instant_checkout" or "price_negotiation."
  2. Contextual Handshake via MCP: The agent establishes a session using the Model Context Protocol (MCP), which provides a standardized way for the LLM to access the merchant’s product catalog. This step replaces traditional web scraping with a structured JSON-RPC exchange, ensuring the agent receives 100% accurate product specifications.
  3. Identity and Permission Verification: The merchant system requests a cryptographic proof of identity from the agent, often utilizing Decentralized Identifiers (DIDs) or OAuth-based agent tokens. This verification ensures the agent has the legal authority to bind its human principal to a financial contract and operates within predefined budget limits.
  4. Dynamic Offer Generation: The ACP server generates a machine-readable offer based on the agent's specific query, incorporating real-time variables like geographic shipping costs, loyalty discounts, and current stock levels. This offer is signed with a private key to prevent tampering during the agent's reasoning process.
  5. Autonomous Settlement: The agent accepts the offer by submitting a secure payment token (such as a virtual credit card or a blockchain-based stablecoin settlement). The ACP gateway validates the payment, triggers the fulfillment workflow, and returns a cryptographically signed receipt that the agent stores in its long-term memory for the user's records.

What to look for

Evaluating an ACP implementation requires a focus on technical interoperability and the security of the autonomous transaction flow.

FAQ

How do I expose my product catalog to ChatGPT and Claude via MCP? Exposing a catalog requires the deployment of an MCP server that acts as a bridge between your database and the LLM. This server implements a set of "tools" that the AI can call, such as search_products or get_inventory_levels. By hosting this server and registering it within the agent's environment, the LLM gains the ability to query your live data directly. This method is significantly more reliable than traditional SEO, as it provides the model with structured JSON data rather than requiring it to parse unstructured HTML.

How do I publish an agent-card.json or llms.txt for my brand? Publishing these files involves placing them in the /.well-known/ directory of your primary domain. The llms.txt file is a markdown-based summary designed to give LLMs a high-level overview of your site's purpose and key endpoints. The agent-card.json is a more technical manifest that defines the ACP version you support, your public encryption keys, and the specific API paths for agent interactions. These files act as the "robots.txt" for the AI era, signaling to crawlers and agents how they should interact with your commerce functions.

What is the difference between MCP, ACP, UCP, and A2A for agent commerce? MCP (Model Context Protocol) is the general transport layer for AI-to-app communication. ACP (Agent Commerce Protocol) is the specific set of rules for buying and selling within that layer. UCP (Universal Commerce Protocol) is an older term often used for cross-platform retail data, while A2A (Agent-to-Agent) refers to the specific communication between a buyer agent and a seller agent. In a standard transaction, an agent uses MCP to talk to a store, follows ACP rules to negotiate, and may interact with other A2A protocols to coordinate delivery or financing.

Does ACP require the use of cryptocurrency or blockchain? ACP is payment-agnostic and does not strictly require blockchain technology. While many agentic systems prefer stablecoins or programmable money for instant settlement, the protocol is designed to work with traditional "Fiat-to-API" gateways like Stripe or Adyen. The primary requirement is that the payment method must support programmatic authorization, allowing the agent to complete the transaction without a human manually entering a CVV code or performing a 3D-Secure biometric check at the moment of purchase.

How does ACP handle returns and customer service for AI purchases? Returns in an ACP-compliant environment are handled through "Reverse-ACP" flows. The agent that performed the purchase retains a digital "Proof of Purchase" token. If the product is defective or incorrect, the agent can initiate a return request by presenting this token to the merchant's ACP return endpoint. The protocol defines standard status codes for "return_initiated," "refund_processed," and "exchange_offered," allowing the AI to manage the entire post-purchase lifecycle on behalf of the human consumer.

Is ACP compatible with existing e-commerce platforms like Shopify or Magento? Compatibility is typically achieved through a middleware layer or a dedicated plugin that translates ACP requests into the platform's native GraphQL or REST API calls. While these platforms were not originally built for autonomous agents, their robust API architectures make them ideal candidates for ACP integration. Most modern implementations involve a "Headless" approach where the ACP server sits alongside the web frontend, drawing from the same inventory and pricing logic but serving it in a format optimized for LLM consumption.

Sources

Published by AirShelf (airshelf.ai).