WebMCP Generator: make your site callable by AI agents

The WebMCP Generator reads a page and generates the files that let AI agents discover and act on your site: potentialAction JSON-LD, MCP discovery files (/.well-known/ai-catalog.json plus a Server Card), the legacy /.well-known/mcp.json, robots.txt additions for AI crawlers, ai-agents.txt and an implementation guide.

Updated 2026-11-21
PlanFree (2 runs) and up
UsesAbout 2 monthly credits a run
TimeAbout 1 minute
Open WebMCP Generator →

Step by step

  1. Open WebMCP Generator #

    Find it under Schema & Entities, Generate, in the left sidebar.

  2. Enter a page URL #

    Use a page with actions: search, booking, quote, contact or checkout.

  3. Generate #

    The page is read, the site type detected and the files generated.

  4. Hand the files to your developer #

    Each file comes with where to put it.

  5. Verify #

    Check agent readiness with the Agent Readiness tool after publishing.

Why it matters#

AI agents increasingly complete tasks for people: finding a product, booking an appointment, requesting a quote. They prefer sites that describe their actions in machine-readable form. These files are how a site says "here is what an agent can do here, and how".

What it checks#

From the page it detects the site type and available actions, then generates:

File Purpose
potentialAction JSON-LD Describes the actions the page supports
/.well-known/ai-catalog.json and Server Card MCP discovery for agent clients
/.well-known/mcp.json The legacy MCP manifest some clients still read
robots.txt additions Rules for AI crawlers
ai-agents.txt A plain-text guide for agents
Implementation guide Where each file goes and how to test it

Reading the results#

Review each action it found. If an action you offer is missing, the page does not expose it clearly (for example a booking form loaded only by script).

WebMCP Generator readiness and capabilities
Example dataThe readiness verdict, what MCP discovery finds on the site today, and the agent capabilities the package would make live.

Implementing the fixes#

  1. Add the potentialAction JSON-LD to the page.
  2. Publish the discovery files at the paths given.
  3. Merge the robots.txt additions with your existing file rather than replacing it.
  4. Test with Agent Readiness.

Worked example#

A hotel generates WebMCP files from its booking page.

The generator detects a lodging site and two actions: search availability (the date form) and contact. It does not find book a room, because the booking engine loads in an iframe from another domain after the page opens.

The hotel publishes the potentialAction JSON-LD with the two actions, the discovery files under /.well-known/ and ai-agents.txt, following the implementation guide. For booking, it asks its booking engine provider for a direct booking URL with date parameters, adds it as a ReserveAction, and regenerates.

Best practice#

Only declare actions that work for an unauthenticated visitor, and keep the files in sync with the site.

Troubleshooting

The URL could not be fetched #

The app saysCould not fetch that URL, check it is publicly accessible.

Check the page loads in a private window and allow CiterankBot if a firewall is involved.

An action you offer is missing #

The page does not expose it in the HTML: the form loads by script, sits in an iframe from another domain, or opens in a pop-up. Give the action a plain URL or form in the page, then regenerate.

The .well-known files return 404 after uploading #

Some servers and hosts hide folders starting with a dot. Check the files load in a browser at the exact paths in the implementation guide, and ask your host to allow /.well-known/.

Not listed here? See Common problems for limits, checks, plans and pages that cannot be read.

Questions

Do I need this? #
It matters most for sites where agents will act: booking, quoting, ordering, searching a catalogue. For a brochure site it is lower priority than content and schema.
What is potentialAction? #
Schema.org markup describing an action a page supports (search, order, reserve), with the URL template to perform it.
Does it use checks? #
No. It uses about 2 monthly credits a run.

Something here does not match what you see in the app? Tell us and we will fix the guide.