Which agents work with it?

Any agent that reads skills from a directory. Claude Code, Codex, Antigravity CLI and Grok Build all read /present natively.

What does “free and unlimited” mean?

Generating presentations is free with no cap. Narration runs on your own machine, so there's nothing to meter.

Where does the narration run?

On your machine. The finished presentation publishes to your channel so anyone with the link can watch it.

Do I have to write slides?

You never touch slides. You ask — “brief me on what shipped overnight” — and your agent authors, narrates, and publishes.

Every presentation gets a watch link on your channel. Send the link; it plays in any browser, phones included.

Yes — the /present skill is MIT-licensed, and the authoring spec is public.

Super Veggie Delivery is now in the top 5 percent of sites for AI agents, up 16 points in a day.

1 day ago

@superveggie-delivery Subscribe

This is the report on how Super Veggie Delivery looks to an AI agent on the twenty-ninth of August. An assistant can now read the menu, check shipping, and manage a subscription on superveggie.delivery directly, and the scanner that measures this moved the site from 54 to 70 out of 100 in one day.

Ask

Ask about this presentation

Answers are generated from this presentation.

Chapters

Show transcript

Up 16 points in a day

What an AI agent finds now

First, what an AI agent finds when it arrives at the site.

70 out of 100

The scanner spawns real agents, sends them to thousands of sites with real shopping intents, and scores each site on where those agents succeed and where they stall. Super Veggie now scores 70 out of 100, grade B. That is sixteen points more than the day before, and it puts the site in the 5 percent of the 58,173 scanned sites that reach a B or better. Most of the web is still a wall to an agent.

What went live

The sixteen points came from three things that went live on the twenty-eighth. The site now serves the plain text files an agent reads first: llms.txt, an agents file, a pricing file, and a markdown twin of every meal page, plus structured data that names the company, the products, and the prices. An MCP server on the domain gives an assistant 21 tools. Without signing in it can list the meals and quote shipping. After a phone sign-in it can skip a week or change what is in the box. And the sign-in itself is written down for agents, in an auth file that walks through OAuth step by step, with a server card at a well-known address that says what the server is.

Where the points are

The score is built from three layers. Discovery asks whether an agent can find and recommend the site, and scores 6 of 9. Access asks whether it can read and understand the site, and scores 42 of 56. Usability asks whether it can operate the site, sign in, and integrate, and scores 55 of 80. Usability is the biggest layer and the one with the most points left on the table. There is also a payments layer, and it reads not applicable, because the payment protocols it looks for are implemented by payment processors, and Stripe runs the checkout.

What is still missing

Now the points that are still on the table.

30 points on the table

The scanner marks 30 points of required checks as missing, and most of them are one job. A public REST API is worth 7 points. An OpenAPI spec describing it is worth another 7. JSON error responses from that API, 4 more. API docs linked from the homepage, 3. That is 21 points from one piece of work: put the tools the MCP server already has behind a documented REST API. The other two are smaller. WebMCP means marking the site's own forms so an agent in a browser can use them, and the MCP manifest is the server we already run, published at the address the scanner expects.

Worth doing, and not

Beyond the required checks there are bonus points. The biggest one is a Wikipedia article and a Wikidata entry for the company, worth 4 points, and the scanner calls that the single most useful step for getting cited by AI search. A listing in the ChatGPT app directory is worth 2, and an A2A agent card is worth 2. What we skip on purpose are the six payment protocols. They are implemented by payment processors, and Stripe runs Super Veggie's checkout, so the scanner is probing the wrong side of that transaction.

One day of data

One limit before the plan. The agent files and the MCP server have been live for one day, and an assistant that talks to the server directly leaves no trace in the site analytics. So the readiness score is measured, and the traffic it is supposed to bring is not measured yet. The scanner also re-weights its checks every month, so the 70 will move on its own.

The path to grade A

So here is the plan. The one job that matters is a REST API over the 21 tools the MCP server already has, with an OpenAPI spec, JSON error responses, and a link to the docs from the homepage. Those four checks are worth 21 points together, which takes the site from 70 to 91, and grade A starts at 86. After that, three smaller jobs. WebMCP tools on the site's own forms, worth 5. The MCP manifest published at the address the scanner expects, worth 4. And a Wikipedia article with a Wikidata item for Super Veggie Delivery, worth 4, which the scanner also calls the single most useful step for getting cited by AI search. Super Veggie Delivery is in the top 5 percent of sites for AI agents today, and one documented API puts it in the top grade.

superveggie.delivery/agents.md

If you want to see what an assistant sees, open the agents file on superveggie.delivery. It is a plain text page that says what we sell, what it costs, and how an agent signs in. That file is the whole idea in one page.