agentreadyaudited 2026-10-14

grade C · 100% coverage

Report for

github.com

https://github.com/webmachinelearning/webmcpvisit site ↗

Find
friction
Understand
clear
Act
friction

Survival: 75; Can an agent read and act at all?
Superpower: 30; WebMCP tools exposed
If WebMCP added: 79; Headline with WebMCP at full marks

Do this first

ranked by severity × category weight

Machine-readable structurehigh

No structured data (JSON-LD)

Add a JSON-LD block describing the page's primary entity.

Show the fix

One script tag turns guesswork into facts. Start with WebSite/Organization, then the page type (Product, Article, FAQPage…).

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "name": "github.com",
  "url": "https://github.com/",
  "potentialAction": {
    "@type": "SearchAction",
    "target": "https://github.com/search?q={query}",
    "query-input": "required name=query"
  }
}
</script>

Reference ↗

02
Form semanticshigh
Controls outside a <form>
Wrap controls in a <form> with an explicit submit control.
Show the fix
Agents submit forms via standard HTML — not div onclick handlers.
<form action="/submit" method="post"> …fields… <button type="submit">Submit</button> </form>
03
Agent discoverymedium
robots.txt blocks 1 AI agent
Distinguish training crawlers from user agents.
Show the fix
Block GPTBot/CCBot if you like; allow ChatGPT-User and Claude-User so people's agents can act for them.
User-agent: GPTBot Disallow: / User-agent: ChatGPT-User Allow: / User-agent: Claude-User Allow: /
Reference ↗

Category breakdown

Agent discovery
1 of 5 discovery signals present.
40
weight 18% · confidence high
Machine-readable structure
Agents must infer what this page is from prose alone.
65
weight 18% · confidence high
Access & renderability
Agents can read this page without a browser.
100
weight 18% · confidence high
Form semantics
Forms are labelled and typed — an agent can fill them.
75
weight 16% · confidence high
Actionability
Actions are real links, buttons and forms.
98
weight 15% · confidence high
WebMCP capabilitysuperpower
Partial WebMCP support — the tools aren't fully declared yet.
30
weight 15% · confidence medium

Find

friction

Can an agent locate this page and its entry points?

Agent discovery

40/100

1 of 5 discovery signals present.

mediumrobots.txt blocks 1 AI agentBlocked for this page: Bytespider. Blocking training crawlers is a legitimate choice, but user-driven agents (e.g. ChatGPT-User, Claude-User) act on behalf of a person and cannot complete tasks here.▶

Evidence

robots.txt · Disallowed tokens: Bytespider

Fix

Distinguish training crawlers from user agents.

Block GPTBot/CCBot if you like; allow ChatGPT-User and Claude-User so people's agents can act for them.

User-agent: GPTBot
Disallow: /

User-agent: ChatGPT-User
Allow: /

User-agent: Claude-User
Allow: /

Reference ↗

mediumNo sitemap foundAgents that plan multi-step tasks use the sitemap to find the right page without crawling.▶

Evidence

sitemap.xml · https://github.com/sitemap.xml not found and no Sitemap: line in robots.txt

Fix

Publish /sitemap.xml and reference it from robots.txt.

Most frameworks generate one; Next.js supports app/sitemap.ts.

<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
  <url><loc>https://github.com/</loc></url>
</urlset>

Reference ↗

lowNo feed or API description linkedNothing on the page tells an agent where structured data or an API lives.▶

Evidence

raw-html · No <link rel=alternate type=application/rss+xml> or OpenAPI link

Fix

Link your API description or a feed from <head>.

A single <link> makes the machine entry point discoverable without guessing URLs.

<link rel="service-desc" type="application/vnd.oai.openapi+json" href="https://github.com/openapi.json">
<!-- or -->
<link rel="alternate" type="application/rss+xml" title="Updates" href="https://github.com/feed.xml">

Reference ↗

lowNo MCP discovery filesNothing under /.well-known tells an agent whether this site exposes an MCP server. Only relevant if you run one — but if you do, this is how agents find it.▶

Evidence

well-known · https://github.com/.well-known/oauth-protected-resource, mcp.json, mcp/server-card.json not found

Fix

If you host an MCP server, publish /.well-known/oauth-protected-resource.

Required by the MCP authorization spec; clients use it to locate your authorization server.

{
  "resource": "https://github.com/mcp",
  "authorization_servers": ["https://auth.github.com"],
  "scopes_supported": ["read"],
  "bearer_methods_supported": ["header"]
}

Reference ↗

1 thing already working

✓llms.txt present — A curated, LLM-friendly summary of the site is available. (Adoption by agents is still uneven — this is a bonus, not a foundation.)

Understand

clear

Can it tell what the page is without guessing?

Machine-readable structure

65/100

Agents must infer what this page is from prose alone.

highNo structured data (JSON-LD)Without schema.org markup an agent must infer what this page is — a product, an article, an organisation — from prose. That inference is where agents make mistakes.▶

Evidence

raw-html · No script[type=application/ld+json] found

Fix

Add a JSON-LD block describing the page's primary entity.

One script tag turns guesswork into facts. Start with WebSite/Organization, then the page type (Product, Article, FAQPage…).

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "name": "github.com",
  "url": "https://github.com/",
  "potentialAction": {
    "@type": "SearchAction",
    "target": "https://github.com/search?q={query}",
    "query-input": "required name=query"
  }
}
</script>

Reference ↗

lowNo canonical URLLanguage and canonical URL help agents choose the right version of a page and parse its text correctly.▶

Evidence

raw-html · lang=en; canonical=none

Fix

Declare lang and a canonical link.

Two attributes; prevents duplicate-URL confusion and wrong-language parsing.

<html lang="en">
<head>
  <link rel="canonical" href="https://github.com/webmachinelearning/webmcp">

2 things already working

✓Semantic landmarks present — main, nav, header/footer
✓Clean heading outline — One h1, 58 headings, no skipped levels.

Access & renderability

100/100

Agents can read this page without a browser.

3 things already working

✓Response headers allow agent access — No noindex/noai directives or bot-management headers detected.
✓No CAPTCHA markup detected — No reCAPTCHA, hCaptcha, or Turnstile widgets found in the raw HTML.
✓Content is readable without JavaScript — Primary content appears in the raw HTML; no JS-dependence heuristics fired.

Act

friction

Can it complete the task — book, buy, submit?

Form semantics

75/100

Forms are labelled and typed — an agent can fill them.

mediumControls missing a name attribute1 control lacks name — without a name the value cannot be submitted or described in a tool schema.▶

Evidence

raw-html · div:nth-of-type(2) > div:nth-of-type(1) > div:nth-of-type(2) > div > span > input @ div:nth-of-type(2) > div:nth-of-type(1) > div:nth-of-type(2) > div > span > input

Fix

Add a name attribute to every control.

Agents need stable field keys for submission and tool schemas.

<label for="field">Go to file</label>
<input id="field" name="field" type="text">

highControls outside a <form>inputs outside a <form> cannot be submitted by an agent without executing scripts▶

Evidence

raw-html · div:nth-of-type(2) > div:nth-of-type(1) > div:nth-of-type(2) > div > span > input @ div:nth-of-type(2) > div:nth-of-type(1) > div:nth-of-type(2) > div > span > input

Fix

Wrap controls in a <form> with an explicit submit control.

Agents submit forms via standard HTML — not div onclick handlers.

<form action="/submit" method="post">
  …fields…
  <button type="submit">Submit</button>
</form>

mediumMissing submit affordance1 control is not inside a form with a submit button — an agent cannot send the values.▶

Evidence

raw-html · div:nth-of-type(2) > div:nth-of-type(1) > div:nth-of-type(2) > div > span > input @ div:nth-of-type(2) > div:nth-of-type(1) > div:nth-of-type(2) > div > span > input

Fix

Add a submit button inside the owning form.

Agents need a standard submit control to complete form actions.

<form action="/submit" method="post">
<label for="field">Go to file</label>
<input id="field" name="field" type="text">
  <button type="submit">Submit</button>
</form>

1 thing already working

✓All controls are labelled — 1 control with accessible names.

Actionability

98/100

Actions are real links, buttons and forms.

medium6 links have no destinationThese anchors do not expose a usable navigation target.▶

Evidence

raw-html · div:nth-of-type(1) > div:nth-of-type(2) > div > span:nth-of-type(1) > a
```
<a class="Link--inTextBlock" href="">Reload</a>
```
raw-html · div:nth-of-type(1) > div:nth-of-type(2) > div > span:nth-of-type(2) > a
```
<a class="Link--inTextBlock" href="">Reload</a>
```
raw-html · div:nth-of-type(1) > div:nth-of-type(2) > div > span:nth-of-type(3) > a
```
<a class="Link--inTextBlock" href="">Reload</a>
```

raw-html · div > div:nth-of-type(1) > nav > ul > li:nth-of-type(2) > a

<a href="#" aria-current="page" class="prc-components-UnderlineItem-7fP-n"><span data-component="icon"><svg data-component="Octicon" aria-hidden="true" focusab…

raw-html · div > div:nth-of-type(1) > nav > ul > li:nth-of-type(3) > a

<a href="#" class="prc-components-UnderlineItem-7fP-n"><span data-component="icon"><svg data-component="Octicon" aria-hidden="true" focusable="false" class="oc…

Fix

Replace the non-native control with the matching native HTML element.

Native controls are keyboard-operable, appear in the accessibility tree, and are what agents (and WebMCP) can target.

<!-- Use a button for an in-page action -->
<button type="button">Open cart</button>
<!-- Or give navigation a real destination -->
<a href="/cart">View cart</a>

low1 focusable elements have no roleA tabindex without a role tells agents something is interactive but not what it is.▶

Evidence

raw-html · div#repos-split-pane-content

<div class="content__cw9d2I" data-selector="repos-split-pane-content" id="repos-split-pane-content" tabindex="0"><div style="--spacing:var(--spacing-none)" cla…

Fix

Replace the non-native control with the matching native HTML element.

Native controls are keyboard-operable, appear in the accessibility tree, and are what agents (and WebMCP) can target.

<!-- before -->
<div tabindex="0">Details</div>
<!-- after -->
<button type="button">Details</button>
<!-- Or add role="button" with keyboard handling -->

1 thing already working

✓Primary action 'Secret protectionStop leaks before they start' is a real <a> — The page's primary action is directly invokable from its raw HTML.

WebMCP capability

30/100

Partial WebMCP support — the tools aren't fully declared yet.

1 thing already working

✓modelContext.registerTool referenced in page script — document.modelContext.registerTool is referenced in an inline script — registration can't be verified without executing the page, so this is credited as a strong signal rather than a confirmed tool.

Fetched Wed, 14 Oct 2026 19:24:26 GMT in 1.5s · HTTP 200 · 8 discovery files checked · rendered DOM: skipped

Rendered-DOM checks are not run in v1; JavaScript dependence is estimated from the raw HTML.