So far in this series the theme has been making your store legible to AI: clean feeds and structured data so agents can read you, and protocols and payments so they can buy from you. Feeds and pages have one thing in common though: they are static. They describe your store at the moment they were generated. The next step closes that gap. Instead of handing an agent a document to read, you hand it a set of live tools it can call: search products, check stock, add to cart, answering with real data at the instant it asks. No scraping, no stale snapshot, no confident guess about a product you delisted last week.
Why tools beat feeds
A feed and a tool answer the same questions in opposite ways. A feed says, in advance, here is everything I sell and everything about it, frozen at export time. A tool says, ask me anything you need and I will answer with what is true this second. For a fast-moving catalog the difference is the difference between accurate and embarrassing. Prices change, stock runs out, variants sell through. An agent working from last night's feed can recommend a sold-out size at yesterday's price. An agent calling a live check stock tool simply cannot. Tools also let the agent act, not just look: adding to a cart, applying a code, or starting a return are verbs a feed can never offer.
The control story is just as important. When an agent scrapes your pages it takes whatever it can parse, however it interprets it. When it calls a tool, it can only do the exact things you chose to expose, and it gets back exactly the data you chose to return. You are not opening your store to a bot roaming in the dark. You are handing it a clean, labeled front desk with a fixed set of buttons. That is why "expose tools" is a control move as much as a discovery move.
MCP and WebMCP, side by side
Both let an agent call your tools. The difference is where the agent is when it calls them. That single distinction decides which one fits a given job.
MCP: the server side (the agent connects to you over the network)
The Model Context Protocol is an open standard Anthropic introduced on November 25, 2024. It gives AI systems one common way to connect to external tools and data, replacing a custom integration per app with a single shared interface. In commerce terms, you (or your platform) run an MCP server that publishes tools like search products, check inventory, or create cart. Any agent that speaks MCP can connect to it from wherever it runs and call those tools directly. This is not theoretical: Shopify's Storefront MCP and Stripe's MCP server are already live, exposing real store and payment actions to agents today.
Best when: an agent needs to reach your catalog or systems from anywhere (a chat assistant, a backend workflow, another company's app), independent of any open browser tab.
WebMCP: the in-browser side (the agent is already on your page)
WebMCP takes the same idea and moves it into the browser. Instead of connecting to a server you host, your web page uses JavaScript to declare tools to the AI agent running in the visitor's own browser, in their current signed-in session. The agent can then call, say, your on-page search or add to cart tool with the shopper's context (their login, their cart, their location) already present. It is being developed in the W3C Web Machine Learning Community Group, jointly driven by Google and Microsoft, and it is a proposal, not a finished web standard.
Best when: the action belongs in a live, logged-in browser tab, where the shopper's session and context are exactly what makes the tool useful.
The honest standardization status
This is where hype and reality need separating, because the two technologies are at very different stages.
MCP is real and in production. It launched as an open-source standard in late 2024, adoption spread quickly across the AI ecosystem, and through 2025 its stewardship has been moving toward more neutral, cross-industry governance, the kind of shift that reduces single-vendor lock-in and reassures merchants that the protocol is not one company's private lever. Live commerce servers exist. If you want live agent tools you can rely on now, MCP is the mature path.
WebMCP is early. It is a proposal in a W3C Community Group, explicitly under active discussion and subject to change, not a ratified standard. Support is Chrome-first: it is available through a Chrome origin trial (beginning with Chrome 149) and a local developer flag, with the specification led by Google Chrome and interest from Microsoft. That is a genuinely promising signal, two major browser makers engaging early, but it is a bet, not a platform. Prototype on it, watch it, do not rebuild your checkout around it this quarter.
Do this now
You almost certainly will not implement either protocol by hand. You will get them through your platform and a partner. So the practical work is knowing what to turn on and what to ask for:
- Ask your platform if it already exposes an MCP server. If you are on Shopify, Storefront MCP is a real, documented capability. Find out which tools it exposes and whether it is switched on for your store.
- Decide which tools you actually want agents to have. Search and check-stock are low-risk and high-value starting points. Actions that move money or data deserve more thought and the same merchant-of-record guardrails from the checkout post.
- Prototype WebMCP behind the Chrome origin trial. A simple on-page search products tool is enough to learn the model and be ready. Keep it experimental; do not make it load-bearing.
- Keep your feeds and structured data clean anyway. Tools do not replace agent-readable pages and feeds; they sit on top of them. Agents that cannot call a tool still fall back to reading you.
If you run Shopify and want the concrete, hands-on version of the WebMCP side of this, our walkthrough of WebMCP and Shopify shows what switching it on actually looks like for a store. Next in the series we turn from capability to strategy: the disintermediation risk, the real danger that, as agents become the gatekeepers, brands get flattened into price-and-spec line items, and what merchants do to keep their differentiation, their data, and their customer.
Frequently asked questions
What is MCP (the Model Context Protocol)?
MCP, the Model Context Protocol, is an open standard that lets an AI system connect to external tools and data through one common interface instead of a custom integration per app. Anthropic introduced it on November 25, 2024 and released it as an open-source project. For a store, MCP is the server side: you run (or your platform runs) an MCP server that exposes tools an agent can call, such as search products, check stock, or add to cart. The agent calls those tools and gets back live, structured data straight from your systems, rather than scraping your pages. Live commerce servers already exist, including Shopify's Storefront MCP and Stripe's MCP server.
What is WebMCP, and how is it different from MCP?
WebMCP is a proposal to let a website expose tools to the AI agent running inside the user's own browser, using JavaScript on the page itself. MCP is the server side: an agent connects to a server you host over the network. WebMCP is the in-browser side: your web page declares tools that the browser's agent can call in the visitor's current session, with their login and their context already present. They are complementary. MCP is the better fit when an agent needs to reach your catalog from anywhere; WebMCP fits when the action happens in a live, signed-in browser tab. WebMCP is being developed in the W3C Web Machine Learning Community Group and is a proposal, not a finished web standard.
Is WebMCP a web standard I can rely on today?
Not yet. WebMCP is a proposal in the W3C Web Machine Learning Community Group, not a ratified web standard, and it is under active discussion and subject to change. Support is Chrome-first: it is available through a Chrome origin trial (beginning with Chrome 149) and a local developer flag, with the specification led by Google Chrome. Treat WebMCP as an early bet worth understanding and prototyping, not a stable platform to build your business on this quarter. MCP, by contrast, is already in production use with real commerce servers, so if you want live agent tools today, MCP is the more mature path.
Why would a merchant expose tools to AI agents at all?
Because tools give the agent live, accurate data instead of a guess. When an agent scrapes your pages it can misread prices, miss stock changes, or confidently recommend a product you stopped selling. When it calls a tool you expose, it reads your real catalog, your real prices, and your real inventory, and it can only do the specific actions you allow. You decide which tools exist and what data they return, so you get accuracy and control at the same time. As AI assistants become a shopping surface, the stores that expose clean tools are the ones agents can actually use, while everyone else stays a page to be scraped.
Sources & further reading: Anthropic: Model Context Protocol, WebMCP spec (W3C Web Machine Learning Community Group), WebMCP in Chrome, and Shopify Storefront MCP. MCP's launch date (November 25, 2024) and WebMCP's Chrome origin-trial availability (from Chrome 149) are per the primary sources above; WebMCP remains a Community Group proposal under active change, and MCP governance continues to evolve, so treat both status notes as current-as-of-writing and confirm details before you build.