agentradar

Check whether a site (and, for app builders, the underlying product) is visible and usable to AI agents: LLM crawlers, shopping/browsing agents, and MCP clients.

It is deliberately narrow. AgentRadar does not check general site health, performance, or SEO. Every check answers one question: can an AI agent discover, read, or use this?

Usage

npx agentradar scan <url>
# or, with Bun:
bunx agentradar scan <url>

Add --json for machine-readable output:

npx agentradar scan example.com --json

What it checks

20 checks grouped by outcome, not protocol name:

• Can agents find your site - robots.txt, AI bot rules, sitemap.xml, Link headers, DNS-AID
• Can agents read your content - markdown negotiation, llms.txt, Content Signals
• Can agents reach your app - apple-app-site-association / assetlinks.json deep-link association
• Can agents trust and transact with you - API Catalog, OAuth discovery + protected resource, Auth.md, MCP Server Card, A2A Agent Card, Agent Skills, UCP, x402, Web Bot Auth, MPP

The composite score is a tier-weighted pass ratio (error 4, warning 2, notice 1); notice-tier checks are emerging standards that carry low weight, so adopting them lifts a site from a B into an A. Any failing error-tier check caps the grade at C. Checks tagged inferred confirm only that a signal is published, not that it is enforced, and are excluded from the score.

Full check reference and rationale: https://github.com/codixus/agentradar#readme.

License

MIT