Agent Native Auditor

Web app to audit whether a site follows key agent-native web principles.

What it checks

• /.well-known/agents.json availability and schema quality
• /llms.txt availability and structure signals
• Homepage discovery hints (<link rel="alternate">, meta name="agent-capabilities")
• Accept: text/markdown support
• robots.txt and sitemap.xml coverage
• JSON-LD structured data presence
• WebMCP-related HTML tooling signals
• Optional sitemap-based multi-page crawl coverage

Each check returns:

• pass/warn/fail status
• evidence from HTTP response and parsed content
• recommendation for fixing gaps
• weighted points that roll up into a total score
• runtime duration and crawl metadata
• optional per-page sitemap sample diagnostics

Audit modes

• single: homepage-focused audit
• sitemap: crawls sitemap.xml URLs (or the sitemap URL listed in robots.txt) and computes coverage metrics across sampled pages

Sitemap mode crawls all discovered sitemap URLs by default.

The UI also includes:

• recent run history (stored in localStorage)
• score delta against previous run for the same domain
• per-page signal matrix for sitemap samples
• JSON report download for sharing/debugging

Stack

• Next.js App Router
• TypeScript
• Tailwind CSS

Run locally

npm install
npm run dev

Open http://localhost:3000.

API

POST /api/audit

Request body:

{
  "url": "https://example.com",
  "mode": "sitemap"
}

mode supports single or sitemap.

Optional: include maxPages to manually limit sitemap crawl breadth for very large sites.

One-click Vercel deploy

1. Push this repo to GitHub.
2. Set NEXT_PUBLIC_GITHUB_REPO_URL in your Vercel project (see .env.example).
3. Use this clone URL pattern for one-click deploy links:

https://vercel.com/new/clone?repository-url=https://github.com/mattlgroff/agent-native-auditor&project-name=agent-native-auditor&repository-name=agent-native-auditor

The app includes a Deploy This Auditor button that uses NEXT_PUBLIC_GITHUB_REPO_URL when provided.

The included vercel.json sets Next.js framework defaults and standard install/build commands.

Reference specs and resources

• WebMCP proposal: https://github.com/webmachinelearning/webmcp
• agents.json spec: https://github.com/wild-card-ai/agents-json
• markdown-to-agents reference: https://github.com/vercel-labs/markdown-to-agents