Food DTC Let the AI Crawlers In. Agents Still Can't Buy.

Six of the ten food and beverage brands I checked let every AI crawler straight through - no blocking, no 403s. It bought them nothing. They sit on the same 3-of-9 protocol floor as every other vertical I have run. Opening the door to the crawlers turns out to be a long way from being ready for the agents, and food DTC is the clearest proof of it yet.

Same nine-signal check from the beauty batch, where every protocol links to its spec. Food and beverage lands like this, and each brand links to its live result.

Eight of the ten are on Shopify and inherit the same three signals - a storefront MCP, UCP and an llms.txt - at the platform level, per Shopify's agentic commerce docs. immi is the exception at 2 of 9, missing its llms.txt. Every self-hosted signal - WebMCP, A2A, ANP, AP2, OpenAPI - is empty across all ten. Nobody built anything of their own.

Crawler access is not agent readiness

This is the part food DTC makes plain. OLIPOP, Fly By Jing, Ghia and Jeni's each let all twenty AI crawlers through and still score 3 of 9. Omsom lets all twenty through and scores 0. An open door means a crawler can read your pages; it says nothing about whether an agent can build a cart and check out, which is what these nine protocols actually describe. That gap between readable and buyable is the one the checkout mechanics piece traces at the transaction layer. Most of food DTC has cleared the easy half and not started the hard one.

Two brands reached 4 of 9, on the same borrowed point

Magic Spoon and Death Wish Coffee are the only two above the floor, and they got there the way the apparel leaders did: a Wikidata entity the checker matched to their domain. It is a notability signal, free to register, and the rest of the category has not bothered. No food brand self-hosted an actual protocol to earn a point - the fourth check is borrowed from being well known, not built.

Custom stacks score zero, open doors or not

Omsom and Spindrift both run custom stacks instead of Shopify, and both score 0 of 9 - no inherited MCP, UCP or llms.txt, no Wikidata match. The tell is in the crawler column: Omsom is wide open (20 of 20) and Spindrift blocks all twenty, yet the protocol score is identical. Crawler posture and agent readiness move on separate tracks, and off-platform you start both from scratch. What the build actually takes, platform by platform, is in the platform requirements piece.

What food and beverage DTC should fix first

1. Do not mistake open crawlers for readiness. Most of you already let the bots in, which is good and also just the floor. An agent still needs a protocol to transact, and none is present.
2. Register a Wikidata entity. It is the one point the two leaders hold and the rest do not, and it costs nothing if your brand has any press to cite.
3. Put back the llms.txt Shopify ships. immi lost its own and sits at 2 of 9 for it - the cheapest point on the board to give away.
4. Self-host one real signal to clear 4. A WebMCP tool surface or an agent card is unclaimed across all ten, so shipping one puts you first in the category.
5. On a custom stack, build the whole checklist yourself. Omsom and Spindrift inherit nothing; off Shopify every signal is yours to ship.

Beauty capped at 3, apparel reached 4 on borrowed fame, and food and beverage shows that a door thrown wide open to crawlers still leaves you at the same floor. Reachable is not buyable. Agentic commerce is already here, and the only question is whose product an agent can actually add to cart and buy - the way that compounds is in the pillar guide.

Run the agent-readiness checker on your store and see whether an agent gets anything past your open door.