Skip to content

Agent-readiness index

Most sites are invisible to agents.

Wasl scores whether a business is legible to AI agents, then generates the MCP server that makes it legible. The score is a deterministic function over evidence — models propose capabilities and explain findings, but every claim cites the markup that justifies it, and anything uncited is refused.

What an agent receives

<!doctype html>
<html lang="en">
  <head>
    <title>Marsa Properties</title>
    <script defer src="/main.8f2c91.js">
  </head>
  <body>
    <div id="root"></div>
    <noscript>You need to enable
      JavaScript to run this app.</noscript>
  </body>
</html>

0 products · 0 prices · 0 identifiers

What a person sees

Waterfront Apartments in Dubai Marina

3 listings · verified by the brokerage · updated daily

2 Bedroom, Marina Gate Tower 1 — AED 2,850,000

1 Bedroom, Bluewaters Residences 4 — AED 2,100,000

3 Bedroom, JBR Sadaf 6 — AED 4,400,000

3 listings · 3 prices · 0 identifiers

Same URL, same moment. The left pane is the raw HTTP response; the right is the DOM after hydration. Wasl measures the ratio between them — 0.02 here — and that single number says more about agent-readiness than any other signal on the page.

Scan a site

Leave it unticked and the scan still runs. It stops before writing any artifacts and asks again.

Read-only · 0.5 req/s · robots-respecting · what the crawler does

100 points, six axes

  • 1Machine-Readable Identityrobots, sitemap, llms.txt, canonicals15
  • 2Structured Data Coverageschema.org entities and their validity20
  • 3Capability Exposurespecs, manifests, stable discovery URLs25
  • 4Content Extractabilityserver-rendered vs hydration-only15
  • 5Transactional Integritystable IDs, structured pricing, forms15
  • 6Agent Governance & Safetyagent-aware terms, injection surface10

A scan that reaches fewer than 8 pages, or that is robots-blocked on more than 30% of them, reports LOW CONFIDENCE with the grade band suppressed. A confident-looking score on thin evidence is worse than no score.

What it refuses to do

Score with a model

No model call returns a number. The rubric is a pure function over evidence, in a package that cannot import the model layer — enforced by walking the import graph in CI.

Ship an uncited capability

A capability without evidence cannot be constructed. The critic rejects five named failure modes, and every rejection is published beside the successes.

Generate a write tool

Read-only tools only. State-changing capabilities are detected and reported, never emitted — generating a booking tool for a site we do not control is how this becomes an incident.