What is an agents.md file?

An agents.md file is a plain-text or Markdown document at the root of your domain that tells AI shopping agents how to interact with your store: where your machine-readable product feed lives, what your checkout and returns rules are, how prices and availability are expressed, and which automated actions you permit. It is the agent-commerce equivalent of robots.txt plus a quick-start guide, giving an agent the context it needs to recommend and buy your products correctly.

How is agents.md different from llms.txt?

llms.txt is a general map of your site's most useful content for language models, oriented toward discovery and answering questions. agents.md is transaction-oriented: it focuses on the commerce surface, pointing agents to your product feed, pricing and availability, and the rules for buying. Many stores publish both, with llms.txt helping an agent find you and agents.md helping an agent buy from you.

Do I still need product schema if I have agents.md?

Yes. agents.md is a pointer and a rulebook, not a replacement for structured data. Agents still rely on Product, Offer, and AggregateRating schema, plus a machine-readable feed such as products.json, to read exact price, currency, availability, GTIN, and variants. agents.md works best when it directs an agent to clean, complete structured data rather than standing in for it.

Is agents.md an official standard?

It is an emerging convention rather than a ratified specification, similar to how llms.txt grew bottom-up. Because it is a simple file you control, the cost of adopting it is low and the downside is minimal even if conventions shift. Keep it small, accurate, and pointed at structured data that follows established schemas so it stays useful as the ecosystem settles.

agents.md: making your store buyable by AI shopping agents

A small file at the root of your domain is becoming the difference between a store an AI agent can recommend and one it can actually buy from. Here is what to put in it.

For Shopify and DTC founders and marketers · Hatchloop

For most of the last decade, the buyer on the other side of your storefront was a human with a browser. That is changing. AI assistants are starting to do more than answer "what is the best moisturizer for oily skin." They compare options, check whether something is in stock, read your return policy, and in a growing number of cases complete a purchase on a person's behalf.

When the shopper is an agent, your beautiful product page is mostly noise. The agent does not care about your hero video. It needs three things fast: what you sell, what it costs right now, and whether it can buy it. An agents.md file is how you hand an agent that context directly, instead of hoping it scrapes the right pieces off a rendered page.

What agents.md actually is

An agents.md file is a plain-text or Markdown document you place at the root of your domain (for example, yourstore.com/agents.md). Think of it as the agent-commerce sibling of robots.txt and llms.txt: a short, predictable place an automated buyer can look to understand how to work with your store.

Where llms.txt is about discovery — helping a model find your most useful pages and answer questions about you — agents.md is about transaction. It tells an agent where your machine-readable product data lives, how your prices and availability are expressed, what the rules of buying are, and which automated actions you allow. It is a pointer and a rulebook, not a brochure.

Worth being honest: this is an emerging convention, not a ratified standard. But it is a file you fully control, it costs almost nothing to publish, and the downside if conventions shift is minimal. That is a good risk profile for something this early.

Why this matters now

The shift toward agent-led buying is happening at exactly the moment most stores are unprepared for the discovery that precedes it. In our 2026 audit of 22 DTC brands, the picture was consistent and uncomfortable.

What the Hatchloop DTC audit found

Median AI-visibility score: 52 out of 100. No brand scored above 64.
32% had zero product schema on their pages.
27% were already behind a direct competitor on AI-discoverability.
91% had no MCP endpoint — no programmatic way for an agent to query the store.
Median agent-commerce readiness, by contrast, sat at 86 out of 100 — the rails for buying exist; the data to ride them does not.

Full methodology and results: State of AI Visibility in DTC.

Read those two medians together and the gap jumps out. The infrastructure to complete an agent purchase is largely in place (86/100), but the structured information an agent needs to choose and trust a product is missing or thin (52/100, a third with no schema at all). Agents can transact. They just cannot read most stores well enough to want to.

That is the opening. A large and growing share of buying journeys now start with an AI assistant rather than a search results page. When an agent has to pick between two comparable products and one exposes clean, machine-readable price, availability, and policy data while the other forces guesswork, the legible one wins by default. agents.md is the front door to making your store the legible option.

What to include in your agents.md

Keep it short and accurate. A bloated file that drifts out of date is worse than a small one that is current. Here is a practical skeleton.

# agents.md — yourstore.com

## About
Brief one-line description of what you sell and your shipping regions.

## Product data
- Product feed: https://yourstore.com/products.json
- Structured data: Product + Offer schema on all product pages
- Currency: USD (prices include/exclude tax — state which)

## Availability & pricing
- Availability and price are authoritative in products.json and Offer schema
- Updated in near-real-time; do not cache price beyond 15 minutes

## Buying
- Checkout: standard Shopify checkout / agent-checkout endpoint (if any)
- Returns: 30-day window, see https://yourstore.com/returns
- Supported regions: US, CA, UK

## Permissions
- Allowed: read catalog, read price/availability, add to cart
- Not allowed: scraping customer reviews for resale, bulk export

## Contact
- Support: support@yourstore.com
- MCP endpoint (if live): https://yourstore.com/mcp

The sections that earn their place:

A pointer to your feed. Name the exact URL of your products.json or equivalent. Do not make an agent crawl to find it.
Pricing and availability ground truth. State which source is authoritative and how fresh it is. Agents need to know whether they can trust a cached price or must re-check before buying.
Buying rules. Checkout path, return window, supported regions, shipping basics. These are the questions a human reads your footer for; agents need them in one place.
Permissions. What automated actions you welcome and which you do not. This is also where you signal that you want agent traffic — many stores accidentally look hostile.
An endpoint, if you have one. If you run an MCP endpoint, link it. Given that 91% of audited brands had none, simply having one is a differentiator.

How it pairs with products.json and machine-readable price

This is the part founders get wrong: agents.md does not replace structured data. It directs agents to it. The file is the map; the territory is your feed and your schema. Both have to be good.

Shopify already exposes a products.json endpoint that lists products, variants, and prices in JSON. That is a real head start, but the default output is often incomplete for agent use — missing or inconsistent GTINs, vague availability, no clear currency or tax treatment. The pairing that works looks like this:

products.json gives the agent the full catalog: variants, SKUs, prices, stock state, in one fetch.
Product and Offer schema on each page gives the same facts in a form agents and search engines already understand, with price, priceCurrency, and availability as explicit fields rather than text an agent has to interpret.
agents.md ties them together: it tells the agent both sources exist, which is authoritative, and how often they change.

The non-negotiable across all three is machine-readable price and availability. "In stock" buried in a badge image is invisible. "availability": "https://schema.org/InStock" and a numeric price with a priceCurrency are not. An agent will not guess your price from a strikethrough graphic, and it will not risk recommending an out-of-stock item it cannot verify. When that data is missing — as it was for roughly a third of the brands we audited — you are not just lower in the ranking, you are unreadable.

Get the three layers consistent and the payoff compounds. The same structured data that makes you buyable by an agent also makes you more citable in AI answers, which is where a growing share of discovery now happens.

A realistic starting sequence

You do not need to boil the ocean. In order of leverage:

Add Product and Offer schema everywhere it is missing — start here if you are in the third of stores with none. This single move lifts both AI visibility and agent-readiness.
Audit your products.json for complete prices, currency, availability, and identifiers.
Publish agents.md pointing at both, with clear buying rules and permissions.
Add llms.txt so agents can discover you in the first place.
Consider an MCP endpoint once the data layer is clean — it is the differentiator almost nobody has yet.

None of this is hype-driven future-proofing. It is the same work that already improves how you show up in AI-generated answers, done once, in a way that also makes you transactable. The stores that do it now will be the defaults agents reach for when the buyer on the other side is no longer human.

See where your store stands

Most DTC brands score in the low 50s on AI visibility and do not know it. Find your number, then close the gap.

Run a free AI Visibility Check Product Schema Generator llms.txt Generator

Want the fixes prioritized and done? The $99 AEO Audit + Fix List turns your score into a ranked, store-specific action plan — schema gaps, feed issues, and a ready-to-publish agents.md included. See the full benchmark in the State of AI Visibility in DTC report.