AAgentVisible

Methodology · scanner 0.4.0

Scores should measure relevant capability, not reward decorative files.

AgentVisible separates discoverability, readability, actionability, trust and reliability, then scores only the checks that apply to the selected or detected website profile.

AgentVisible scoring methodology

Dimensions

DimensionWhat it measures
DiscoverableWhether agents can find useful machine-facing resources.
ReadableWhether agents can obtain clean, structured content.
ActionableWhether declared public actions have appropriate machine-readable interfaces.
Trust & accessWhether identity, permissions and authentication are explained.
ReliableWhether capabilities remain available over time. Phase 0 does not yet measure this dimension.

Applicability

A missing check is not automatically a failure. Informational sites are not penalised for lacking OpenAPI, MCP, Agent Skills or WebMCP. API profiles are assessed for API discovery and access guidance; agent-service profiles are assessed for agent tool discovery and access guidance. Present positive evidence is retained even when it is optional for a profile. Service, ecommerce and booking currently share the universal discoverability/readability baseline; those labels prepare the policy model without inventing action requirements.

AgentVisible also fingerprints the hosting platform (Shopify, WooCommerce, WordPress, Magento, BigCommerce, Squarespace, Wix, Webflow) from public homepage evidence. Platform detection feeds platform-aware checks: standard Shopify storefronts expose a live Storefront MCP server at /api/mcp, and a verified tools/list response is credited as real Actionable evidence. Absence of a platform endpoint is never penalised — headless storefronts legitimately do not expose it.

Use the optional type parameter to declare intent: https://agentvisible.io/scan?url=example.com&type=api. Without it, AgentVisible classifies from homepage content first: software-application markup, or a sign-in/app entry point combined with pricing, plans, features or docs pages, infers SaaS; commerce markup (schema.org Product/Offer types, add-to-cart or cart/checkout paths, or a recognised shop platform) infers ecommerce; lodging/reservation markup or booking paths infer booking. SaaS evidence outranks commerce markup because software companies routinely use commerce-style markup for their own product. Only without content signals does protocol evidence classify: MCP/skills/WebMCP evidence infers agent-service, API Catalog/OpenAPI evidence infers API, and everything else is treated as informational. Content inference is deliberately conservative and can always be corrected with the type parameter.

Headline score

The headline score is the weighted share of applicable checks that pass, normalised to 100. Not-applicable checks are removed from the denominator. Dimension scores use the same rule. Warnings do not currently receive partial credit. The headline is profile-specific: Level 4 is reserved until Actionable, Trust & access and Reliable are all measured and strong; Phase 0 reports otherwise excellent sites as Level 3.

Evidence and versioning

Every result includes scannerVersion and schemaVersion. Evidence is collected from the live public origin at scan time. A passing file check validates minimum resource-specific structure rather than accepting any HTTP 200 response. Network and scanner failures are currently reported within failed checks and should be interpreted alongside their evidence field.

Independence

Paid or managed status cannot change a score or ranking. Future public profiles will show freshness, scanner version and correction routes.

Data handling

Phase 0 does not persist public profiles. Scan reports are cached in process for 60 seconds to reduce repeat outbound requests; the cache is lost on restart. Public profile persistence and its retention controls will be introduced separately.