• 목록
• 답변
• 글쓰기
• 게시판 리스트 옵션
  • 수정
  • 삭제

What's agent readiness? Agent readiness is how legible and usable a website, API, or MCP server is to autonomous AI agents. A human customer studying a hero on a touchdown page is ok without structured knowledge. An agent that hits the same URL wants express machine-readable indicators: a sitemap it could possibly crawl, an llms.txt at the root, OpenGraph plus JSON-LD on each page, predictable rate-restrict headers, an OpenAPI spec for any public endpoint, and (more and more) a discoverable MCP server card or AP2 manifest. Without these signals the agent guesses, fails halfway, or offers up. The Open Agent Readiness scanner right here measures all of that towards a public spec so you'll be able to fix the gaps before brokers start showing up at your site. Each scan teams checks into seven categories. A failing category drops the general rating; a green category contributes weighted factors. The class breakdown lives in the public leaderboard and within the JSON returned by the v1 API. Sitemap.xml, robots.txt, llms.txt, llms-full.txt, MCP server card, AP2 manifest, OpenAPI spec at a predictable path.

If an agent can't find your endpoints by crawling, the remainder of the categories don't matter. OpenGraph tags on each page, JSON-LD structured information (Organization, Product, FAQPage, BreadcrumbList), canonical URLs, and meta-description protection. Determines whether your content surfaces in AI search (Perplexity, ChatGPT search, Claude search) and will get cited appropriately. Rate-limit headers (RateLimit-Limit, RateLimit-Remaining, Retry-After), proper 401/403 semantics, idempotency-key help on Post/PUT, CORS configured for cross-origin agent calls. What the endpoint actually exposes. For MCP servers: variety of instruments, sources, prompts. For Rest APIs: documented endpoints, schemas, instance responses. For sites: structured product or pricing data. Whether the endpoint can take an agent's money: x402 micropayments, MPP (the Tempo / Stripe Machine Payments Protocol), AP2, and ACK-ID for pay-per-call access, and whether or not prices are marketed in machine-readable form. We probe every individually, so an MPP-only endpoint (which answers with WWW-Authenticate: Payment, not x402) is detected too. The bar Coinbase and Stripe each endorsed. Response-time consistency, schema stability throughout requests, deterministic error mental illness shapes, content material-type honesty (returning JSON when the docs say JSON).

Identity verification (DNS records, signed manifests), phrases-of-service URL discoverability, ai.txt or similar bot-coverage recordsdata, and uptime track report from prior scans. Agentic site visitors is not hypothetical. ChatGPT search, Perplexity, Claude, Gemini Deep Research, Cursor, Devin, and a protracted tail of vertical brokers all crawl the general public net and name APIs as a part of normal request dealing with. By Q1 2026 agentic site visitors on B2B SaaS websites already accounts for 5-15% of all non-bot requests, and the curve is steeper than the human cellular transition was. Sites that ship readiness indicators get cited in AI answers, picked up by agent activity plans, and routed to in agent-driven discovery flows. Sites with out them get politely skipped. The leaderboard above tracks who's already ahead. Paste a URL, hit scan. The scanner makes a sequence of HEAD and GET requests, parses HTML, fetches sitemaps, probes effectively-identified paths (/llms.txt, /.well-identified/mcp.json, /.nicely-known/ap2.json, /openapi.json), and runs spec-outlined checks towards every response. Non-2xx responses are categorised (skip vs fail) so a lacking-but-non-obligatory file would not penalise you a similar as a misconfigured-required one.

Results render as the rating ring above plus a paste-ready fix immediate for every failure that you can drop into Claude or Cursor and have it carried out in minutes. The identical spec runs over MCP for AI agents that want to self-audit, and over Rest at /api/v1/agent-readiness. Missing llms.txt. The one most common hole. Add a plaintext file at the foundation itemizing your key URLs and a one-paragraph site description. JSON-LD on the homepage only. Engines like google already settle for this; brokers need it on product, pricing, docs, and individual content material pages too. Generate from a shared helper, not hand-rolled. Rate-restrict headers missing on API responses. Even in case your limits are generous, the lack of headers makes agents back off conservatively. Add RateLimit-Limit, RateLimit-Remaining, RateLimit-Reset to each authenticated route. MCP server with no card. In case you publish an MCP server, drop the JSON card at /.nicely-known/mcp.json with identify, description, version, transport, tools array.

Without it the server is invisible to discovery. 402 endpoint with out published pricing. Return the worth within the 402 response physique in the format the x402 spec defines. Otherwise paying brokers cannot funds the decision. Cloudflare's product is about blocking agent traffic; this scanner is about welcoming it. The 2 solve opposite problems. Should you actively want to be discoverable to agents (most websites should, by 2026), you want the readiness audit; if you'd like to keep brokers out of premium content material, Cloudflare is the reply. They compose effectively: gate paid endpoints behind Cloudflare, make the remainder of your surface readiness-compliant. Full breakdown on the devoted comparison web page. Is the scanner free? Yes. Public scans are free with no signup required. The /api/v1/agent-readiness endpoint prices 1 credit per call (Free plan includes 1,500 credits monthly, loads for occasional checks). Paid agent access through x402 can also be supported for autonomous brokers with no API key. Do you store the URLs I scan?