Agentic commerce readiness: what to do before August 31 (and what to skip)

This is part 3 of a 3-part series on agentic commerce standards. Part 1 maps what’s actually shipped; part 2 fact-checks the claims in circulation. This part is the to-do list.

The first two parts of this series were about separating signal from noise. This one is about the signal only: what a brand should actually do, in the next 60 days, about the agent layer that exists — and what it can safely skip.

One date organizes the whole post. On August 31, 2026, per Shopify’s changelog, the original Storefront MCP cart tools stop being maintained. It’s the second half of a migration that already happened to the catalog tools in June. The platform’s agent layer is consolidating onto UCP, and anything pointed at the old endpoints is running out of road.

First, the reframe: agents already call your store

The most common misconception Lumio hears from brands is that agentic commerce is something to opt into later. On Shopify, it is already running. Per Shopify’s documentation, every store has an MCP endpoint that exposes catalog search, cart building, and policy answers — no authentication required, no merchant action taken. Every store also serves a default /agents.md discovery file and a /.well-known/ucp protocol manifest.

That changes what “readiness” means. The question is not whether an AI agent can call your store. It’s what the agent gets back when it does — and that answer is a direct function of your product data.

When an agent calls search_catalog on your store, the quality of the response is your catalog’s quality: whether products carry complete prices, availability, variants, identifiers, and descriptions rich enough to match how shoppers actually phrase requests. When it calls the policy tool with “what’s the return window?”, the answer is only as good as the policy content you’ve published. The endpoint is Shopify’s; everything it serves is yours.

The checklist

1. Find anything built on the old cart tools — before August 31. If your team or your agency built anything against get_cart or update_cart on /api/mcp — a custom shopping agent, a chatbot integration, internal tooling — it needs to move to the UCP Cart MCP tools at /api/ucp/mcp before the maintenance window closes. Note the behavioral change while migrating: per the changelog, the new update_cart uses PUT semantics — each request replaces the cart’s full state rather than patching it. If nobody built anything custom, this item costs you one question to your developers.

2. Look at what an agent sees. Fetch your own store’s /agents.md and /.well-known/ucp. If your /agents.md is the Shopify default, it reads nearly identically to millions of other stores’ — customizing it via the theme template is cheap and puts your brand’s actual voice and priorities in the file agents read first. Keep expectations honest: discovery files are hygiene, not a ranking lever. The heavy lifting happens in the catalog data behind the endpoint.

3. Fix the catalog data the endpoint serves. This is the highest leverage item and the least glamorous. Agent-visible catalog responses inherit every gap in your product data: missing GTINs, absent prices on variants, thin or duplicated descriptions, images without context. The same structured-data completeness that determines how your products reach AI answers through indexes and feeds determines what the MCP endpoint can say about them. If your products are invisible to agents on one path, they’re usually degraded on all of them, and for the same underlying reason.

4. Make your policies answerable. Agents ask pre-purchase questions — returns, shipping windows, sizing, warranties — through the policy and FAQ tools. Published, current, specific policy content is what makes those answers accurate. A missing return policy used to cost you a support ticket; on the agent path it can cost you the sale, because the agent moves on to a store that answered.

5. If you’re on BigCommerce: opt in, labeled as what it is. BigCommerce’s Storefront MCP is first-party and available to every live store via the Early Access toggle — and per BigCommerce’s own docs it is in beta. Opting in is low-cost and gets your catalog into the agent path early; treating it as battle-tested infrastructure would be premature. Guest checkout flow only, for now.

What to skip: the WebMCP work

Some agencies are already selling “WebMCP readiness.” Parts 1 and 2 covered the status in detail; the operational summary is short. WebMCP is an origin trial in one browser, with no Intent to Ship, one formal opposition (Apple’s WebKit, which proposed setting the proposal aside entirely), one undecided engine vendor (Mozilla), and exactly one named consumer agent (Gemini in Chrome, inside the trial, alongside a developer inspector extension). It also isn’t MCP — nothing you build for it transfers to the server-side layer above, and nothing above transfers to it.

There is no scenario this quarter where WebMCP work beats catalog work on return. The rational posture is watchful neglect. Three events would change that, and they’re all public:

• Chrome files an actual Intent to Ship (the earliest estimate in the record points at roughly Q1 2027)
• The W3C workshop Apple proposed for the TPAC timeframe (late October 2026) produces a redirection — or a resolution
• WebKit’s position changes, or a major commerce platform ships WebMCP tools in production

If one of those fires, revisit. Until then, every hour spent making a page “WebMCP-ready” is an hour not spent on the endpoint agents are calling today.

The honest summary

Three findings from this series, compressed:

1. The agent layer that matters is server-side and already live. Shopify mints an MCP endpoint per store under UCP; BigCommerce is in open beta. What agents get from those endpoints is your product data, served back to you.
2. The browser-layer noise is mostly unsourced. The claims that WebMCP shipped, that major brands committed, that cross-browser support lands in 2026 — none traces to a primary source, and Apple’s formal opposition goes unmentioned in nearly all of it.
3. The deadline is real and specific. August 31, 2026, for the old cart tools. Migration, catalog completeness, and policy content are the work between now and then.

Lumio scores what AI agents actually see in your catalog — the structured data, the completeness gaps, the fixes ranked by impact — and is building the same agent’s-eye check for the UCP/MCP endpoints this series covered. If you want to know where your store stands before August 31, talk to Lumio.