The endpoints appeared quietly on stores in early May. The official word came on May 28, when Shopify's developer changelog announced template customization for three of them. As of this writing, every Shopify store we checked serves five agent-facing surfaces — we verified these live on major storefronts this week, all returning clean 200s:
| Endpoint | What it is |
|---|---|
/llms.txt | "Agent Instructions" — introduces the store to AI crawlers and shopping agents |
/llms-full.txt | The long-form variant (serves the agents.md content by default) |
/agents.md | The operating manual: UCP discovery, checkout rules, agent etiquette |
/.well-known/ucp | JSON merchant profile — UCP versions, services, MCP endpoint |
/sitemap_agentic_discovery.xml | Discovery index for agents |
This isn't a beta you opt into. Open yourstore.com/llms.txt right now — it's almost certainly there.
What the defaults actually say
Open any Shopify store's /llms.txt and the first thing you notice: it's written to the agent, not about the store. The default opens with "Agent Instructions" and tells personal shopping assistants to install Shopify's Shop skill (shop.app/SKILL.md) — which handles cross-store catalog search, price and discount discovery, buyer-approved checkout through Shop Pay, and order tracking.
Read that twice. Shopify is steering every AI agent that visits any of its storefronts toward transacting on Shopify's own rails. The store description, collections, and contact details are in there too — but the headline content is a recruitment pitch to the agent.
The UCP profile at /.well-known/ucp declares the protocol versions the store speaks (2026-04-08 stable alongside 2026-01-23) and exposes the store's shopping service over MCP — with checkout gated behind explicit human approval and per-IP rate limits. That's the Universal Commerce Protocol shipping as a platform default, not a pilot you sign up for.
What you can customize — officially, since May 28
Per the changelog, three template files in your theme (Online Store → Themes → Edit code → Templates) control the markdown surfaces:
templates/agents.md.liquid— controls/agents.mdand acts as the fallback for both llms pathstemplates/llms.txt.liquid— controls/llms.txtonlytemplates/llms-full.txt.liquid— controls/llms-full.txtonly
The fallback chain: specific template → your agents.md.liquid → Shopify's generated default. Your template replaces the default wholesale — there's no merging. The UCP profile is generated from your store configuration and has no template override. Full reference: Shopify's agents.md.liquid documentation.
Should you customize? Audit first — then upgrade the parts only you can write.
Shopify's default is the same template on every store. The agent mechanics are solid — keep those — but the default cannot know what makes your catalog worth recommending. That's the layer to add: a real store description (who you are, what you sell, who it's for), your hero categories with a one-line positioning each, the products you're known for, and the policy facts agents ask about — shipping, returns, guarantees. An agent deciding between three stores it has never seen ranks on exactly this information.
The method matters because a custom template replaces the default wholesale — there's no merging. So: copy the default content as your template's base, keep the Shop-skill and UCP mechanics intact, and add your brand layer on top. Don't hand-write the structure from scratch — the llms.txt generator drafts the merchant layer from your live site in about a minute. And because the feature is young (content shipped weeks before documentation), date your template and recheck the defaults occasionally; this surface is evolving.
Beyond the brand layer, the stronger override cases: multi-region and Markets stores (verify what the root-level defaults expose for locale and currency before an agent quotes the wrong market's price), regulated or age-restricted catalogs that need conditions stated to agents, and B2B-only merchants. The only stores that should leave the defaults entirely alone are ones that won't commit to maintaining a template — a stale custom file is worse than Shopify's evolving default.
What to do this week
- Read your four markdown-and-JSON surfaces.
/llms.txt,/agents.md,/.well-known/ucp, and the discovery sitemap. You can't improve a surface you haven't read. - Check what agents see beyond the manifest. The manifest invites them in; your PDPs still have to be parseable. Run a product page through the AI Crawler Simulator for the bot's-eye view, and the Agentic Readiness Grader for the checkout-signal audit.
- Markets stores: verify locale and currency in what the defaults expose. The files live at the domain root; confirm your international markets aren't being described in the wrong language or currency.
- Add your brand layer. Copy the default into
templates/agents.md.liquid, keep the mechanics, and add the store description, category positioning, and policy facts only you can write. - Watch the changelog. This area is moving fast — the endpoints appeared, evolved, and got documented inside a single month.
The strategic read
Shopify just settled the "is agentic commerce real?" question for its merchants: the infrastructure is on by default, for everyone, with the Shop skill positioned as the transaction layer. Which moves the competition to the only place it was ever going to live — whether the engines and agents reading these manifests find your products parseable, citable, and recommendable. The manifest gets the agent in the door. The PDP wins or loses the recommendation.
That part was always the real work. Now it has a deadline.
Frequently asked questions
Does my Shopify store already have an llms.txt file?
Should I create a custom llms.txt.liquid template?
What is the difference between llms.txt and agents.md on Shopify?
Does this mean AI agents can check out on my store now?
Ask AI about this
Have your favorite engine explain what Shopify's native llms.txt means for your store.