HoookdBot
HoookdBot is the crawler of Hoookd, a tool that checks how well a website works for AI models and AI agents. It visits a site only when someone asks Hoookd to scan that site.
What it does on your site
- It reads
/robots.txtfirst and follows it. - It fetches at most 25 pages per scan, at most one per second.
- It loads a few key pages in a browser to see what they look like after scripts run.
- It only reads. It never submits a form, creates an account or buys anything.
- Hoookd does not train AI models on what it reads.
What a scan sends to other services
A scan reads your site and keeps what it collects in this deployment’s own database. These are the other services it contacts, and what each is told:
- The MCP registry. The name of your domain goes to
registry.modelcontextprotocol.io, in one or two searches, to see whether an MCP server is listed for it. No page content is sent. - OpenRouter. When this deployment has a language-model key, the text of your key public pages (the homepage and the pricing, features, about, docs, signup and FAQ pages) goes to a language model through
openrouter.ai. The model is asked whether those pages answer basic questions about the product. The name of your product and your domain also go to a model there, which searches the web for pages about the product on other sites. When the pages leave a basic question unanswered, the site’s name, its page titles and descriptions and its prices go to a model that drafts a clearer description for the report. Without a key nothing is sent, and the report says that these two checks did not run. - AI models, when AI visibility is run. Only a signed-in user can run it. Up to seven questions go to two models through
openrouter.ai, each of which answers from its own web search. Five of the questions are what a buyer might ask and do not name your product. Two name it. The answers are then compared with the text of your key public pages by a model there. Nothing in the questions tells a model what to say about your product. - The profiles your site names. A scan requests up to ten of the addresses your site gives as its own profiles elsewhere (the
sameAslist in its structured data and the social links on its homepage), once each, to see whether they exist. Those sites see a request from HoookdBot for a page you link to, and nothing about your site.
How it identifies itself
Direct requests send this user agent:
Mozilla/5.0 (compatible; HoookdBot/1.0; +https://app.hoookd.space/bot)Pages loaded in a browser send the browser’s own user agent with this added at the end:
HoookdBot/1.0 (+https://app.hoookd.space/bot)One check tests how a site answers other crawlers. Those requests carry the other crawler’s user agent with HoookdBot-probe added at the end, so they can always be told apart from the real crawler.
How to block it
Add this to your robots.txt. HoookdBot will then fetch nothing from your site except robots.txt itself.
User-agent: HoookdBot
Disallow: /HoookdBot follows the group that names it. If no group names it, it follows the rules for User-agent: *.
Contact
This deployment has not published a contact address.