Make docs readable for AI agents and test tool routing

Agents and answer engines are visitors too. Open the Docs site ready for AI agents template.

1. Website for AI agents

Select Blog article (server-rendered). The check runs the AI Website Readiness Explorer either on a public URL or — as here — on pasted Page HTML, robots.txt and llms.txt, which works offline and for pages that are not public yet.

The result: AI readiness 94 of 100 (grade A); 3 of 16 AI bots blocked. The three blocked bots are training crawlers; every answer-engine and user-request bot may fetch the page, so Agentlab does not raise a finding for them. Open AI bots to see each bot, its operator and purpose.

Now select App shell (JavaScript only): 36 of 100 (grade F). The high finding explains why: the HTML contains three characters of text and a script tag, and many AI crawlers do not run JavaScript. The table Checks lists the missing pieces (no <h1>, no meta description, no canonical URL, no structured data).

The gate AI readiness score of websites ≥ 80 uses the worst page, so the shell fails it.

2. Routing test set

Select Ten typical agent requests. A Routing test set has one user request per line, optionally followed by => and the expected usefulweb.site tool id:

Is this MCP server safe to connect to my coding agent? => mcp-server-explorer
How much will our chatbot cost per month with 5,000 requests a day? => llm-cost-explorer

Each line is one call to the Prompt-to-Tool Router. The result: 8 of 10 requests reach the expected tool first, 10 within the top 3, with a mean reciprocal rank of 0.9. The table shows the top tools per request and the rank of the expected one, so you can see which wording confuses the router. Download batch results (CSV) exports the table.

Use it when an agent talks to the usefulweb.site MCP server: if a typical request does not reach the right tool, the agent will not find it either.

3. Index the tools an agent sees

usefulweb.site tools for agents is a Tool definitions check in search mode: it lists this site's tools that match mcp agent, with capability tags and links to their definitions. Tick Also search the official MCP Registry to include public MCP servers (needs the tool API to reach the registry).

4. Track it

Website scores change with every deploy. Set a Review by date on each page check and export Review reminders (.ics); the History tab charts the score of each run.