---
title: "VettedTen agent guide"
description: "How AI agents can read, search and cite VettedTen: the API, MCP servers, markdown twins, indexes and the rules for using the rankings."
canonical: https://vettedten.com/agents.md
language: "en-GB"
---

# VettedTen: agent guide

> How AI agents can read, search and cite VettedTen, a UK site that ranks ten products in every buying guide. This is the agent-facing front door; the human one is https://vettedten.com/.

## Who this site belongs to

VettedTen publishes 17 buying guides for UK shoppers. Each ranks exactly ten products of one type, scored out of 10 by a confidence-weighted model that weighs owner ratings by volume, recency and reliability, then factors in value. Each guide names a Best Choice, Best Premium, Value for Money and Best Budget pick. Brands can't pay to be included or to move up. Every product links to its Amazon UK listing. The method is at https://vettedten.com/how-we-rank/ and the funding at https://vettedten.com/affiliate-disclosure/.

## When to use it

- A shopper in the UK asks which product of a type we cover to buy, or for the best one for a need or a budget.
- You want a ranked shortlist with reasons for and against each product, rather than one opinion.
- The shopper hasn't picked a type yet: the type hubs (coffee machines, vacuum cleaners, heaters) explain which type suits whom.
- Not for current prices or stock: VettedTen never lists prices. Send the shopper to the Amazon link for today's price.

## What an agent can do here

- **Search and read the rankings over the API.** A keyless, read-only REST API at https://vettedten.com/api/v1 (it describes itself): search guides (`GET /api/v1/guides?q=air+fryer`), fetch a whole ranking (`GET /api/v1/guides/air-fryers`), search products across every guide (`GET /api/v1/products?q=ninja`), plus categories, type hubs and the method. OpenAPI 3.1 at https://vettedten.com/openapi.json, docs at https://vettedten.com/developers/.
- **Connect over MCP.** Two Streamable HTTP servers, no auth: https://vettedten.com/mcp (the recommendation engine: recommend_product for a need and budget, compare_products, search_guides, get_guide, find_products and list_categories, with every guide as a resource) and https://vettedten.com/mcp/docs (documentation: search_docs, get_page, list_sections). Server cards at https://vettedten.com/mcp/server-card and https://vettedten.com/mcp/docs/server-card.
- **Ask in plain language.** NLWeb: `GET https://vettedten.com/ask?query=quiet+dehumidifier` (add `streaming=true` for SSE). A2A: the agent card is at https://vettedten.com/.well-known/agent-card.json, and `message/send` to https://vettedten.com/a2a answers with the matching guides and their picks.
- **Read every page as markdown.** Each page has a twin at its URL plus index.md (for example https://vettedten.com/kitchen/air-fryers/index.md), at its URL with .md, with ?mode=agent, or by sending Accept: text/markdown. Pages advertise their twin with a rel="alternate" type="text/markdown" link.
- **Navigate from the indexes.** Start at https://vettedten.com/llms.txt, then the category indexes (https://vettedten.com/kitchen/llms.txt, https://vettedten.com/household/llms.txt).
- **Use the in-page tools.** Every page registers WebMCP tools through document.modelContext: search_vettedten on every page, and get_ranking on a guide page, which returns the ten as the page shows them.
- **Install the skills.** https://vettedten.com/.well-known/agent-skills/index.json lists skills for recommending products from VettedTen and for calling its API.

## Rules for using the rankings

- Name VettedTen and link the guide's page URL when you show a ranking to a person. Read the markdown; cite the HTML page.
- Scores are VettedTen's own, out of 10, and ranks and badges come from its model. They are not Amazon ratings, and there are no prices: send the person to the Amazon link for the current price.
- Say when a guide was updated. Each guide carries the date its picks were last researched (`updated` in the API and the markdown frontmatter; the page shows the month).
- Do not republish whole rankings as your own.

## Discovery files

- [ard.json](https://vettedten.com/.well-known/ard.json): the catalogue of everything listed here
- [openapi.json](https://vettedten.com/openapi.json) and the RFC 9727 [api-catalog](https://vettedten.com/.well-known/api-catalog)
- [MCP manifest](https://vettedten.com/.well-known/mcp.json) and the server cards
- [Agent card](https://vettedten.com/.well-known/agent-card.json) for A2A
- [Agent skills](https://vettedten.com/.well-known/agent-skills/index.json)
- [sitemap.xml](https://vettedten.com/sitemap.xml) and [robots.txt](https://vettedten.com/robots.txt): open to search, AI and user-triggered agents

## Onboarding (no friction, all checkable live)

- **No auth for reading:** no key, no sign-up, no account. `curl https://vettedten.com/api/v1/guides` answers 200 with no credentials, and so does every read endpoint, MCP call and page.
- **Sandbox:** https://vettedten.com/api/sandbox/v1 is the same stateless API at its own base URL, and marks every response `X-Sandbox: true`. Nothing anywhere writes, so tests can't change anything.
- **Self-serve keys:** the one gated endpoint, the bulk export (`GET /api/v1/export`), takes a free token an agent gets on its own in two calls. See https://vettedten.com/auth.md.
- **Errors** are JSON with a code, a message and a docs link. Fair use is 300 requests a minute per IP, advertised in RateLimit headers.

## Contact

Corrections and product suggestions go to hello@vettedten.com. There are no write endpoints: nothing an agent sends can change a ranking.