Skip to content

Methodology

Each checkpoint has a stable id, a stage, a strength and an evidence type. The spec part tells what agents need. The scoring policy tells how we score it.

2026-10-30 Aligned with agentready.org spec 2.0

The spec

Find
Agents can reach the site and discover its pages.
Read
Agents can fetch and parse the content without JavaScript.
Act
Agents can finish tasks through forms, APIs, MCP and auth.
Checkpoints Grouped by stage
Strength and evidence How to read a checkpoint
MUST
An AR MUST is required when applicable. A local MUST is an IsAgentReady priority. Neither label proves task success.
SHOULD
Recommended by the mapped AR practice or by our local policy. Check the strength source.
MAY
Optional. Absence alone is not a conformance failure.
Info
Policy information. It has no strength.
Measured study
A study measured the effect on agents.
Scanner data
Scanner data shows adoption or failure.
Recommendation
A standard or expert guidance only.

AR strengths come from agentready.org. Local priorities, including security headers and document titles, come from IsAgentReady policy. The ar_strengths field lists each mapped practice's normative strength. Mapping and priority never change score weights.

agentready.org practices How our checkpoints map to the open spec
  • AR-FIND-01 Allow agents in robots.txt and bot protection MUST

    Robots directives and HTTP responses from the scanner with selected user-agent strings.

    Provider IP access, server logs, every public page and the expected answer are not verified.

    AI crawler directives , HTTP bot accessibility , Meta robots / X-Robots-Tag , robots.txt present , Training crawler blocking

  • AR-FIND-02 Publish the answer as first-party docs SHOULD

    Documentation links, sampled HTML and freshness signals.

    Canonical authority, factual accuracy, completeness and current answers are not verified.

    Developer docs discoverability , Docs pages that answer

  • AR-FIND-03 List pages in a sitemap.xml SHOULD

    Sitemap response, XML structure and robots references where recorded.

    A sitemap found at a guessed path does not prove a robots reference or complete page coverage. Historical scans may lack reference evidence.

    XML sitemap

  • AR-FIND-04 Publish an AI Catalog and link it MAY

    AI Catalog response, resource entries and recorded discovery links.

    A catalog at a guessed path does not prove homepage-head and HTTP Link discovery. Resource completeness is not verified.

    AI Catalog , Link response headers

  • AR-READ-01 Fetchable without JavaScript MUST

    Text volume and structure in the initial HTML.

    Text volume does not verify that the expected answer is present, accurate or complete.

    Server-side rendered content

  • AR-READ-02 Return real HTTP status codes MUST

    Responses to bounded missing-path and accessibility probes, including observed throttling.

    Unobserved throttling behavior and every route are unknown. Historical scans may not distinguish valid redirects or challenge pages.

    Soft-404 hygiene

  • AR-READ-03 Homepage in raw HTML SHOULD

    Initial homepage HTML, semantic structure and recorded navigation links.

    The expected answers and completeness of routes to relevant product, docs and pricing pages are not verified.

    Document title , Heading hierarchy , Server-side rendered content

  • AR-READ-04 Docs pages that answer SHOULD

    Sampled docs status, text volume, headings and code-language markers.

    These heuristics do not verify factual, complete or current answers to user tasks.

    Developer docs discoverability , Docs pages that answer

  • AR-READ-05 Fenced code, language-tagged SHOULD

    Code block language markers on sampled documentation pages.

    Coverage of every snippet and correctness of the declared language are not verified. Historical scans may accept a partial tagging ratio.

    Docs pages that answer

  • AR-READ-06 Well-known discovery files SHOULD

    Independent general discovery-link observations recorded by ia.link-headers (observed.discovery_links), including on API sites.

    API-specific Link validation and successful guessed-path fetches do not prove general discovery links. Historical scans without the independent observation remain unverified; coverage of every relevant page is not verified.

    A2A Agent Card , ACP manifest , API Catalog (RFC 9727) , Link response headers , UCP

  • AR-READ-07 llms.txt SHOULD

    llms.txt response, format and index links.

    Index quality, factual descriptions, destination answers and complete discovery linkage are not verified.

    llms.txt

  • AR-READ-08 JSON-LD SHOULD

    JSON-LD syntax, schema fields and linked entities.

    Valid JSON-LD does not verify parity with visible text, factual accuracy or coverage of key pages.

    Author attribution , Entity linking via @id , FAQPage schema , High-value schema types , JSON-LD present , Organization / WebSite schema , Schema validation

  • AR-READ-09 Markdown mirror MAY

    Recorded Markdown negotiation and alternate mirror responses.

    Equivalent answers in HTML and Markdown, and mirrors for every key page, are not verified. Historical scans may measure negotiation only.

    Content negotiation (text/markdown)

  • AR-READ-11 Pricing on a first-party page SHOULD

    Same-origin pricing-page HTML and recognizable price amounts.

    Amounts do not prove complete plans, a free tier, limits, accurate pricing or an answer to a pricing question.

    Agent content contracts , First-party pricing page

  • AR-ACT-01 Auth an agent can finish SHOULD

    Authentication metadata, PKCE declarations and documented credential hints.

    No key is created and no OAuth, consent or authenticated task is completed. Historical scans may omit API-key documentation.

    Agent auth readiness , Agent content contracts , OAuth discovery , OAuth protected resource

  • AR-ACT-02 Public API with a machine-readable contract SHOULD

    OpenAPI structure, declared responses and bounded read-only endpoint probes.

    Contract completeness, authorization, write operations and successful product tasks are not verified.

    API Catalog (RFC 9727) , API error responses , API rate-limit headers , OpenAPI / API docs , Live OpenAPI verification , OpenAPI actionability quality

  • AR-ACT-03 MCP server SHOULD

    MCP discovery, initialization and listed tools where recorded.

    Useful tool execution, OAuth completion, registry publication and task success are not verified. Historical scans may require an optional server card.

    MCP discovery , Live MCP verification

  • AR-ACT-04 Server cards for pre-connection discovery MAY

    Server-card response and metadata fields.

    Host support and successful pre-connection discovery are not verified; the card convention can change.

    MCP server card quality

  • AR-ACT-05 MCP Apps MAY

    MCP Apps resource and UI metadata hints.

    The view is not rendered in a host and its interactions, permissions and task completion are not verified.

    MCP Apps UI resources

  • AR-ACT-06 SDKs and a CLI SHOULD

    SDK registry links and CLI installation or command hints, when distinguished in observations.

    An SDK does not prove a CLI. Installation, first-call behavior and useful task execution are not verified.

    Agent Skills index , SDKs and CLI

  • AR-ACT-07 WebMCP SHOULD

    WebMCP source annotations and form/tool structure.

    Source inspection does not execute document.modelContext tools or verify signed-in browser tasks.

    Interactive surface coverage , WebMCP declarative API , Declarative form quality

  • AR-ACT-08 Let agents pay MAY

    Payment-protocol metadata and checkout source markers.

    No payment or checkout is executed. Consent, tokens, CAPTCHA handling and purchase completion are not verified.

    ACP manifest , Agent checkout , AP2 , MPP (x-payment-info) , UCP , x402 payments

  • AR-ACT-09 NLWeb ask endpoint MAY

    Bounded NLWeb ask response and schema-shaped result hints.

    An ask endpoint does not prove exposure of the same tool over MCP, answer quality or complete content coverage.

    NLWeb ask endpoint

  • AR-ACT-10 An MCP server or a CLI MUST

    Recorded MCP tool discovery or CLI evidence from the task-access checkpoint.

    A mention or empty interface does not prove a useful task surface. Task execution and whether agents are expected to act on this product remain unknown. Missing optional surfaces do not prove an informational site.

    Agent task access , MCP discovery

Changelog

Version history Current version 2026-10-30
2026-10-30 Clarifies the bounded evidence behind every AgentReady practice. Full and partial mappings describe a relation, not verified conformance. Weights, score roles and maximum points remain unchanged. Historical reports retain their original scores and outcomes; corrected measurements can change new scans. Scanner corrections cover crawler roles, Markdown mirrors, discovery links, sitemap references, status probes, MCP discovery, SDK versus CLI evidence, documentation code tags and authentication documentation. Documentation checks now apply to informational websites as well. No authentication, purchase or product task is executed.
2026-10-22 Added seven checks for the agentready.org practices that the scanner did not cover: pricing page, MCP Apps, NLWeb ask, agent checkout, SDKs and CLI, agent task access and docs pages that answer. All seven are bonus or informational, so core scores do not change. The AR mapping, strength and evidence type are new metadata and do not change scores. Agent task access is MUST in the spec but bonus in the score until a benchmark cycle confirms it. Readiness profile scores can change by a few points, because the new bonus checkpoints count in those profiles. The headline score does not change.
2026-09-02 Made the semantic and discovery checks outcome relevant and retired three checkpoints with weak evidence.
2026-08-29 Revised the FAQPage, interactive surface coverage and OpenAPI quality checks.
2026-08-27 Retired the WebMCP manifest checkpoints because no specification defines a manifest.
2026-08-25 Added soft-404 hygiene, DNS-AID, trust pages, page token budget, API errors, API rate limits and developer docs links.
2026-08-24 First versioned methodology with evidence tiers, score roles and readiness profiles.
Introduced
robots.txt present , AI crawler directives , XML sitemap , llms.txt , Meta robots / X-Robots-Tag , Content freshness signals , Content negotiation (text/markdown) , HTTP bot accessibility , Link response headers , Content-Usage preferences , AGENTS.md agent guide , Agent content contracts , JSON-LD present , Organization / WebSite schema , High-value schema types , Entity linking via @id , BreadcrumbList markup , Schema validation , FAQPage schema , Author attribution , Server-side rendered content , Heading hierarchy , Semantic HTML elements , Landmarks and ARIA , Image alt text , Language attribute , Descriptive link texts , Question-based headings , Form label coverage , Form field autocomplete , Keyboard navigability , Document title , Button accessible names , aria-hidden misuse , Dynamic widget state , WebMCP declarative API , WebMCP manifest , A2A Agent Card , MCP discovery , OpenAPI / API docs , agents.json , Declarative form quality , Manifest tool schema quality , Interactive surface coverage , API Catalog (RFC 9727) , Agent Skills index , x402 payments , MPP (x-payment-info) , UCP , ACP manifest , AP2 , OpenAPI actionability quality , MCP server card quality , Live OpenAPI verification , Live MCP verification , AI Catalog , HTTPS , HSTS , Content-Security-Policy , X-Content-Type-Options , X-Frame-Options / frame-ancestors , CORS , Referrer-Policy , Tool definition security , Auto-submit risk , OAuth discovery , OAuth protected resource , Agent auth readiness

Scoring policy

The score uses weights and points that are separate from the spec. Only core checkpoints change the overall score.

Weights and grades Version 2026-10-30
AI Content Discovery
30%
Content & Semantics
20%
AI Search Signals
20%
Agent Protocols
15%
Security & Trust
15%

A+ 95+ / A 80+ / B 70+ / C 40+ / D 20+ / F 0+

Scoring document (JSON)

Profile checkpoint mapping Which checkpoints count toward each readiness profile