WebMCP Generator

How to generate WebMCP tools

1. 1

   Paste your URL

   Add your site in the dashboard. Nothing to install yet.

2. 2

   The generator reads your pages

   It visits your homepage and up to 5 pages from different sections of your site, found through your sitemap or links, and pulls out the forms, buttons, frameworks and JavaScript globals on each one.

3. 3

   AI writes the tools

   A language model proposes tool definitions for each page: name, description, input schema and executeJs. Near-duplicates are merged, and the rest are ranked by how useful they are to an agent.

4. 4

   You review, test and save

   Each suggestion shows the page it came from. Save it, edit it first, or test it against your real site to see what an agent would get back.

5. 5

   Go live with one script tag

   Paste the snippet once. Saved tools register on the right pages in every major browser, and every call an agent makes shows up in your dashboard.

A run takes 15 to 40 seconds and continues in the background if you close the tab. For the details of what the crawler looks at and why, read how Aigentably generates WebMCP tools from a URL.

What a generated tool looks like

Each suggestion is a complete WebMCP tool definition. This is an example of the shape the generator produces for a store's search page:

{
  name: "searchProducts",
  description: "Search the catalog by keyword. Returns title, price and URL for each match.",
  inputSchema: {
    type: "object",
    properties: {
      query: { type: "string", description: "What the shopper is looking for" },
      maxResults: { type: "number", description: "How many results to return (default 5)" }
    },
    required: ["query"]
  },
  // executeJs: the code that runs in the page when an agent calls the tool
  executeJs: `const res = await fetch('/search?q=' + encodeURIComponent(args.query) + '&view=json')
    ...`
}

Aigentably registers it through document.modelContext.registerTool() on the pages it applies to. The execute code is checked by a security linter before you save. For hand-written patterns, see the WebMCP implementation examples.

Free tools from your schema.org markup

If your pages already carry schema.org JSON-LD for Google, most of a read-only tool is already written. Aigentably turns that markup into tools without calling an AI model, on every plan:

• Product → getProductInfo
• Article / BlogPosting → getArticleInfo
• FAQPage → getFAQAnswer
• Recipe → getRecipeInfo
• Event → getEventInfo
• LocalBusiness → getBusinessInfo
• JobPosting → getJobInfo

More on this in turn your schema.org markup into WebMCP tools.

Generator, hand-written code, or codegen?

These aren't exclusive. Tools you write by hand live next to generated ones, and the generator is told about your existing tools so it fills gaps instead of repeating them.

After generation: test, track, score

• Test on your real site: Run any tool against your live pages before agents can see it.
• See every agent call: Calls, success rate, latency and which agent made them.
• Track your agentic browsing score: Chrome Lighthouse's agentic browsing audit, run weekly or on demand.

Frequently asked questions