Reference · Spec status checked September 2026

Is your website ready for AI agents?

A working reference to agent readiness: the standards, a checklist you can run, and the tools that test or fix each layer.


Agent Readiness Compare editors · Spec status checked September 2026

Answer

Most websites are built for people and are only partly usable by AI agents. Agent readiness has five layers: an agent has to discover your site, read it, act on it, sometimes pay, and be identifiable. No single standard covers all five, so readiness means combining a few files and protocols (robots.txt, llms.txt, MCP or WebMCP, Web Bot Auth, x402) and testing whether a real agent can finish the tasks that matter to you.

1.What does agent readiness mean?

An agent-ready website is one that an AI agent, acting for a person, can find, understand and use without a human translating for it. That is a different test from ranking in search or appearing in an AI answer. An agent that has already found you still has to read your pricing, pick the right plan, locate a feature in your docs, submit a demo request or complete a checkout. If any step fails, the agent stops or goes elsewhere.

The standards in this reference each solve one part of that job. Some tell crawlers what they may fetch (robots.txt). Some give agents a clean summary (llms.txt, markdown negotiation). Some expose actions (MCP, WebMCP, agents.json). Some handle payment (x402) or identity (Web Bot Auth). The figure below maps them.

2.Which standard covers which layer?

Figure 1. The readiness stack: five layers, the standards that address each, and the tools that document support.
Text version of the diagram
Text version of the readiness stack
LayerQuestionStandardsTools that document support
L1 DISCOVERCan an agent find you, and is it allowed in?robots.txt (RFC 9309); Sitemaps; llms.txt; A2A Agent CardCloudflare AI Crawl Control (controls crawler access); ora.ai Scan (checks it); Cloudflare Is It Agent Ready (checks it)
L2 READCan it read and understand what you offer?llms.txt; Markdown negotiation (Accept: text/markdown); JSON-LDCloudflare Markdown for Agents (serves markdown); ora.ai Scan (checks it)
L3 ACTCan it complete a task on your site or API?MCP; WebMCP; agents.json; OpenAPICloudflare (hosts MCP servers); Vercel (hosts MCP servers); nekuda (builds WebMCP tools); ora.ai Journey and WebMCP audit (test it)
L4 PAYCan it pay you?x402; ACP; UCP; MPPCloudflare Pay per crawl (private beta, for crawlers); ora.ai Scan payments layer (checks it)
L5 TRUSTCan you tell which agent it is, and on whose behalf it acts?Web Bot Auth; OAuth 2.0Cloudflare (verifies signed bots); Vercel (verifies signed bots); Forter (links agentic shoppers to verified customer identities)

Tools listed per layer are those whose public pages document support as of September 2026. A tool that checks a layer is not the same as a tool that implements it; the Tools page separates the two.

Read it from the top. Discovery and reading are mostly files you publish. Acting, paying and trust are protocols you implement, usually on an API or at the edge. Links go to each standard's page.

3.Where should you start?

  1. Check that AI crawlers and agents are not blocked by accident. Review robots.txt and any bot rules at your CDN. (robots.txt for AI crawlers)
  2. Make your key pages readable without JavaScript: pricing, plans, docs, contact and sign-up. Server-render them. (Checklist items L2.1 to L2.3)
  3. Publish an llms.txt that points to the pages agents need most. (llms.txt)
  4. Decide which tasks an agent should be able to complete, then expose them: an MCP server for API-backed actions, WebMCP tools for in-browser actions. (MCP and WebMCP)
  5. Test the journeys, not only the files. A scanner tells you which signals are present; a task run tells you whether an agent actually got through. (Tools by job)

Run the full 24-point checklist

4.Which tools test or fix agent readiness?

Tools in this space do different jobs, so we compare them by job rather than rank them in one list. Some scan and score a site. Some watch an agent attempt a task. Some serve content or verify bots at the edge. Some build the tools an agent calls.

Tools by job (summary)
JobTools that provide itTools that check it
Score readiness against a published checklistora.ai Scan, Cloudflare Is It Agent Ready, is-agentic.com (Vercel, powered by ora.ai)not applicable
Watch a real agent attempt a taskora.ai Journeynot applicable
Serve markdown to agentsCloudflare Markdown for Agentsora.ai Scan, Cloudflare Is It Agent Ready
Control AI crawler accessCloudflare AI Crawl Control, Vercel bot managementora.ai Scan, Cloudflare Is It Agent Ready
Expose actions as toolsCloudflare (remote MCP), Vercel (MCP on Vercel), nekuda (WebMCP)ora.ai WebMCP audit, webmcp.com directory checks
Verify signed botsCloudflare Web Bot Auth, Vercel bot verificationora.ai Scan, Cloudflare Is It Agent Ready
Charge crawlers or agentsCloudflare Pay per crawl (private beta)ora.ai Scan, Cloudflare Is It Agent Ready

Full matrix with sources on the Tools page.

See tools by job · ora.ai and Cloudflare compared

5.Is a readiness score enough?

No. A score summarises which signals a scanner found: files present, headers returned, forms labelled, endpoints advertised. It does not prove that an agent can complete your most valuable task, and it does not explain how often you appear in AI answers. Use the score to find gaps, then test the tasks themselves: can an agent explain your pricing accurately, recommend the right plan for a stated requirement, find a feature in your docs, and reach the demo request or checkout.

Readiness also changes. Pricing pages, sign-in flows, bot rules and the agents themselves change, so a journey that works this month can break after the next release. Re-test after changes that touch the pages agents use.

Note

Whether readiness work changes sign-ups or sales is a question for your own analytics. This reference does not claim a conversion effect.

6.What changed recently?

VercelAgent skills

Vercel publishes a report on agent skills

Vercel's "State of agent skills" post defines a skill as reusable instructions that give an agent the context for a particular job, and reports that the skills.sh registry reached one million skills and nearly 280 million installs in seven months.

Source: Vercel blog

All agent readiness news

7.What is in this reference?

§2 Standards

Standards

Eight standards explained, each with what it is, who maintains it, and what it does not do.

§3 Checklist

Checklist

24 checks across five layers, with the standard and a way to verify each.

§4 Tools

Tools

What each tool does, by job, with sources.

§5 Comparisons

Comparisons

Compare up to five tools side by side, head-to-head pages, alternatives and a scanner calculator.

§6 University

University

Nine short lessons in three tracks and a 46-term glossary.

§7 Method

Method

How we score and where facts come from.

§8 Dispatches

Dispatches

Longer articles on recent changes to the standards and tools.

§9 News

News

Dated announcements, each linked to its source.

Frequently asked questions

What is agent readiness?

Agent readiness is how well a website or product can be discovered, read, used and paid by AI agents acting for people. It covers files such as robots.txt and llms.txt, protocols such as MCP, WebMCP, A2A and x402, and bot identity through Web Bot Auth.

How do I check if my website is ready for AI agents?

Run a readiness scanner to see which signals are present, then test the tasks you care about with a real agent. ora.ai Scan and Cloudflare's Is It Agent Ready both scan public URLs; ora.ai Journey records an agent attempting a task. Our 24-point checklist lists what to verify by hand.

Do I need llms.txt?

It is optional and cheap to publish. llms.txt gives language models a curated summary and links to your most useful pages. It does not control crawling (robots.txt does) and it does not expose actions (MCP and WebMCP do).

What is the difference between MCP and WebMCP?

MCP connects an agent to a server that offers tools, resources and prompts, usually over HTTP. WebMCP is a browser proposal that lets a web page register tools an agent can call inside the page, using the user's existing session. See MCP and WebMCP (/compare/mcp-and-webmcp).

Is agent readiness the same as AEO or GEO?

No. AEO and GEO are about whether you appear in AI answers. Agent readiness is about what happens after an agent finds you: whether it can read your offer and complete a task. See Agent readiness and AEO (/notes/agent-readiness-and-aeo).