There is a lot of noise about AI agents that will shop on a customer's behalf, compare options, and check out without the person ever visiting a store. Some of it is hype and some of it is already shipping. I want to cut through it with a practical view: what the emerging protocols actually are, what they need from your Shopify catalog, and which preparations pay off regardless of how the standards shake out.
The protocols, briefly and honestly
A few acronyms are worth knowing, with realistic expectations attached.
ACP, the Agentic Commerce Protocol, is the OpenAI and Stripe effort to let purchases happen inside an AI assistant. The shopper finds a product through the assistant and completes the buy in that surface, with the merchant fulfilling. It is real and rolling out, but availability and merchant onboarding are still early.
MCP, the Model Context Protocol, is a standard for giving AI clients structured access to tools and data. For commerce it means an assistant can query a store's catalog through a defined interface rather than scraping a web page. It is gaining adoption in developer tools and starting to reach commerce.
Agentic checkout is the broader idea these point at: the buy happening where the shopper is talking to the assistant, not on your storefront. Shopify is building toward this too, through its own channels and partnerships.
The honest summary: this is early, the standards are not settled, and you should not bet the business on any single one yet. What you can do is get your catalog into the shape that every version of this rewards, so you are ready when the channel that matters for your store matures.
What all of them need: clean, complete, machine-readable product data
Here is the part that makes this practical. Every one of these protocols, and every assistant that recommends products today, depends on the same foundation: accurate, complete, structured product data. An agent cannot confidently sell what it cannot confidently understand.
That means the unglamorous work is the work that matters. Complete titles and descriptions that state the real attributes. Valid Product structured data with correct price, currency, and availability. Identifiers like GTINs where they exist. Per-variant detail so an agent knows the medium in black is in stock at a specific price, not a vague range. Clear policies, because an agent that is about to transact wants to know your returns and shipping terms.
None of that is speculative. It helps you in Google today, it helps you in ChatGPT and Perplexity today, and it is exactly what an agentic checkout will require tomorrow. The teams that will be ready for ACP or MCP are the ones whose catalog data is already correct, because the protocols are mostly new ways to deliver the same well-structured data.
The few forward-looking moves worth making now
Beyond the foundation, a handful of steps position you specifically for the agentic shift.
Publish an llms.txt index so assistants have a clean map of your catalog. It is cheap and it is the direction the conventions are heading.
Keep your structured data continuously in sync, not just correct once. Agentic surfaces will read your data more often and more literally than a human ever did. Stale availability or a wrong price is far more damaging when a machine is about to act on it. The discipline that matters is freshness, not a one-time cleanup.
Make sure your data is reachable by machines, which means not blocking the AI crawlers in robots.txt and not hiding your product facts behind JavaScript that a fetch cannot see.
If you are on a plan or partner program that offers an agent feed, such as an ACP product feed, the prerequisite for a good one is, again, clean catalog data. The feed is a delivery mechanism; the quality is upstream.
What not to do
Do not rebuild your store around an unsettled standard. Do not pay for a bespoke integration with a protocol whose merchant onboarding is still in flux. Do not treat this as a reason to ignore the fundamentals while chasing the acronym of the month. The fundamentals are the integration. When the standard you care about is ready, connecting a clean catalog to it is the easy part.
Where we land
The agentic commerce shift is coming unevenly and on its own timeline, and the smart preparation is boring on purpose: correct, complete, continuously-synced, machine-readable product data, plus an llms.txt index and open access for the crawlers. Do that and you are ready for ACP, MCP, agentic checkout, and whatever the next acronym turns out to be, because they all want the same thing underneath.
That foundation is precisely what we build for the stores we run, and it is the core of what AgentReady maintains automatically: structured data and an llms.txt index that stay correct as the catalog moves. You do not have to predict which protocol wins. You have to be the store whose data is ready when any of them ask.