Skip to contentRead our launch announcement on LinkedIn

4 min readShopify launches

What Is Actually In Your Shopify llms.txt (We Read 22 Live Stores)

Kratik Agrawal

By Kratik Agrawal

Published Jun 17, 2026 Updated Aug 5, 2026

The opening of a Shopify-generated llms.txt file, headed 'Agent Instructions', showing the store's UCP discovery endpoint, its MCP endpoint, and the rule that checkout requires human approval, alongside scan figures of 15 of 22, 20 of 22 and 2 of 22.

On August 5, 2026 we pulled /llms.txt from 36 direct-to-consumer domains, confirmed 22 of them run on Shopify, and read what came back. Fifteen serve the file. Twenty of those same 22 answer a JSON-RPC call at /api/ucp/mcp that will search their catalog and open a checkout.

Almost none of those merchants asked for any of it. Shopify provisions the agent layer underneath them, and llms.txt is the door it hangs on the front of the building.

Quick answer

llms.txt is a markdown file at yourstore.com/llms.txt that tells AI agents what your store sells and how to transact with it. Shopify generates it for you from live store data. On the stores we checked it names the store's Universal Commerce Protocol endpoints, points at an MCP endpoint agents can call, and states that a human has to approve every payment.

What Shopify puts in your llms.txt

The file opens with a line addressed to shopping agents, then documents how to buy. Allbirds publishes its Universal Commerce Protocol discovery URL, an MCP endpoint at /api/ucp/mcp, a six-step purchase flow from search_catalog through complete_checkout, and the supported protocol versions. It closes with policy links and the rule that agents must not complete payment without explicit buyer consent.

Shopify renders all of it from live store data, so it stays current with no maintenance. A crawler that reads the file gets a working interface description instead of a theme it has to reverse-engineer.

The llms.txt file served at allbirds.com, headed 'Agent Instructions', documenting the store's UCP discovery endpoint, MCP endpoint, and six-step agent purchase flow.
The top of a real Shopify llms.txt. Not a sitemap for robots, a set of operating instructions for buyers that happen to be software. Source: allbirds.com/llms.txt

Does anything actually read llms.txt?

For most of 2025 the honest answer was no, and we said so on this page. That answer has aged badly. The file is now the published entry point to a live transactional interface, and the interface answers.

We sent a tools/list call to the MCP endpoint advertised in these files. Twenty of the 22 stores responded with a working tool list: search_catalog, lookup_catalog, get_product, create_cart, update_cart, create_checkout, complete_checkout, and six more. Two returned a 404. No merchant we have spoken to had switched this on deliberately.

The JSON served at allbirds.com/.well-known/ucp, listing supported Universal Commerce Protocol versions, service endpoints and capabilities.
What sits behind the discovery URL your llms.txt names: the store's machine-readable merchant profile, protocol version 2026-04-08. Source: allbirds.com/.well-known/ucp

20 of 22

Shopify stores answering a live agent call at /api/ucp/mcp

15 of 22

serving an auto-generated llms.txt

2 of 22

naming any AI crawler in robots.txt

Correction

An earlier version of this article said no assistant consumes llms.txt and called it a courtesy map. Our own August 2026 scan shows the file now advertises a working checkout interface on the large majority of Shopify stores. We were wrong and have rewritten the section.

Why free plumbing is not an advantage

Every number above describes something Shopify turned on for every store on the platform at once. Your competitor has the same manifest, the same endpoint, the same 13 tools. A capability that arrives by platform default cannot separate you from anyone.

The gap sits one layer up, in what flows through the pipe. When a buyer agent calls search_catalog, it gets your product titles, your descriptions, your variant names, and your attributes. Thin copy and missing attributes produce a thin answer, and the agent recommends the store that gave it something to work with. That payload is the same product record a feed carries, so the fix is product feed optimization.

Live MCP endpoint: 20 / 22Live MCP endpoint20 / 22UCP manifest: 19 / 22UCP manifest19 / 22Product JSON-LD on PDP: 10 / 12 checkedProduct JSON-LD on PDP10 / 12 checkedllms.txt served: 15 / 22llms.txt served15 / 22Names an AI crawler: 2 / 22Names an AI crawler2 / 22Publishes /agents.md: 1 / 22Publishes /agents.md1 / 22
Kinect scan, August 5 2026. 36 direct-to-consumer domains probed, 26 reachable, 22 verified as Shopify by platform signature. Read-only GETs and one stateless tools/list call per store.

What to fix instead of your llms.txt

Start with the catalog text an agent receives, because that is the payload every one of those 13 tools returns. Fit, fabric, occasion, sizing, what pairs with what. A product page written for a shopper who can already see the photo will read as blank to something that cannot.

Then look at robots.txt. Two of the 22 stores named a single AI crawler in it. The rest are making an access decision by accident, in both directions, and never checking which one they made.

Third, make your policies quotable. Returns, shipping windows, warranty terms. An agent asked whether it can return the shoes will either find a sentence it can lift or hedge, and a hedge costs you the sale.

  • Read your own file at yourstore.com/llms.txt and see what it claims on your behalf.
  • POST a tools/list call to /api/ucp/mcp and read the tool list your store already exposes.
  • Search your own catalog the way an agent would, then read the text it gets back.
  • Decide your AI-crawler policy on purpose and write it into robots.txt.

Where Kinect fits

Kinect's Agent-Ready Storefront covers the layer the platform does not: the catalog text, the structured attributes, and the policies that agents quote. Catalog Enrichment fills in fit, fabric, occasion and pairings, and distributes them to the surfaces where AI shops.

Brands with Kinect see 3–6% more revenue, measured against their own baselines. If you want to see what your store returns when an agent asks it a question today, the UCP Test runs against your live domain and takes about a minute.

Frequently asked questions

What is llms.txt?

llms.txt is a markdown file served at the root of a website that describes the site for AI systems. On Shopify stores it documents the store's agent endpoints and purchase flow, and Shopify generates it automatically from live store data.

Do I need to edit my Shopify llms.txt?

No. It regenerates from store data, so an edit would be overwritten. The content it points at is what you control, and that is where the work belongs.

Will llms.txt get my products into ChatGPT's shopping results?

Not on its own. Shopping results draw on product feeds, crawlable product pages, and reviews. The file makes your store legible to an agent that has already arrived; it does not bring the agent.

Is llms.txt the same as robots.txt?

No. robots.txt sets access rules and crawlers enforce it. llms.txt describes what the site is and, on Shopify, how to transact with it. In a scan of 22 Shopify stores in August 2026, 15 served llms.txt and only 2 named any AI crawler in robots.txt.

How do I check whether my store has a live agent endpoint?

Send a JSON-RPC tools/list call to yourstore.com/api/ucp/mcp. If the store returns a tool list including search_catalog and create_checkout, agents can already transact with it. Twenty of the 22 Shopify stores we checked returned that list.

Related reading

See Kinect on your store

An AI sales rep on your storefront and an agent-ready store behind it — trained on your catalog and policies, live the same day, measured against your own baseline.