Agent readiness checklist summary
What readiness checks cover by layer, the AgentReady spec behind them, and who tests how much of it.
In short
Agent readiness checks cover five layers: find, read, act, pay and trust. The AgentReady spec turns them into 24 identifiers in three stages, four of them MUSTs. The spec scores nobody. Public scanners test very different shares of those identifiers, from about a sixth to all of them by this site's mapping.
What the checks cover
This page summarises what a readiness check looks at. It does not give step-by-step fixes. The five layers come from lesson 1, and each answers one question an agent's task depends on.
| Layer | Question | What the checks look at | Standards |
|---|---|---|---|
| Find | Can an agent find you, and is it allowed in? | Whether robots.txt and network rules let agents through, and whether key pages are discoverable | RFC 9309, Content Signals |
| Read | Can it read what you offer? | Whether the answer is in the initial HTML, whether status codes are accurate, and whether markdown, llms.txt and structured data exist | llms.txt |
| Act | Can it complete a task? | Whether an MCP server, CLI, WebMCP tools or an API description let an agent do the task | MCP, WebMCP, NLWeb, agents.json |
| Pay | Can it pay you? | Whether a machine-readable payment path exists where you charge | x402, UCP, ACP |
| Trust | Can you tell which agent it is? | Whether your edge verifies signed requests and delegates access safely | Web Bot Auth |
The AgentReady spec
AgentReady launched on 19 October 2026 as an open standard for what a product should implement so agents can use it. Its repository README says the spec is written by ora.ai and Vercel, and that the research was conducted by the ora.ai research lab. Ora's launch post lists committee reviewers by area: Ora, Vercel, Mintlify, Auth0, Stripe, Tavily and Elementor. Google and MIT are named as partners there.
Version 2.0, from October 2026, has 24 live identifiers. They sit in three stages, and each carries a strength level of MUST, SHOULD or MAY.
| Stage | Live identifiers | Range |
|---|---|---|
| Find | 4 | AR-FIND-01 to 04 |
| Read | 10 | AR-READ-01 to 09 and 11; AR-READ-10 is retired |
| Act | 10 | AR-ACT-01 to 10 |
The spec has no separate pay or trust stage. Agent payments are AR-ACT-08, a MAY. Verified identity touches AR-FIND-01 through Web Bot Auth.
The four MUSTs
- AR-FIND-01: do not block agent and crawler access in robots.txt or at the network layer.
- AR-READ-01: put the answer in the initial HTML.
- AR-READ-02: return accurate status codes.
- AR-ACT-10: expose an MCP server or a CLI when agents should act on the product.
Everything else is SHOULD or MAY. The MUST, SHOULD, MAY levels reflect what an agent fails without, which practices changed agent behaviour, and which conventions are still emerging.
What version 2.0 changed
- AR-ACT-10, an MCP server or a CLI, became a MUST.
- WebMCP (AR-ACT-07) rose to SHOULD.
- AR-READ-10 was retired.
- New identifiers: AR-FIND-04 (AI Catalog), AR-READ-11 (first-party pricing), AR-ACT-08 (agent payments) and AR-ACT-09 (NLWeb).
The repository recorded these changes in its changelog, last pushed on 21 October 2026.
What the spec does not do
The spec names scoring as a non-goal. It sets no weights, thresholds or badges, and it introduces no manifest format of its own. It scores nobody. Scoring is left to scanners and their methodologies, and the spec points to Ora's. A spec identifier is not a grade, and passing every MUST is not a ranking.
How many identifiers each scanner tests
The counts below are this site's own mapping of each scanner's published checks to the 24 identifiers, read on 4 November 2026. They are not scores and not quality ratings.
| Scanner | Identifiers covered | Note |
|---|---|---|
| Ora | All 24 | 132-check catalog, from robots and bot detection to MCP, WebMCP, x402, UCP and ACP |
| agent-ready.dev | About 15 | The 72-check list leaves out bot protection, the AI Catalog, WebMCP and CLIs |
| Cloudflare | About 11 | Does not test JavaScript-only content, status codes, JSON-LD or code fences |
| Mintlify Agent Score | About 6 | Documentation only; nothing on act or pay |
| SiteSpeakAI | About 4 | Four named areas; no check list is published |
| is-agentic.com | Same checks as Ora | Capability checks count only in the 20-point Recommended pool when a site offers them |
The Scanner Concordance shows how those differences play out on sixty real domains.
Sources
- AgentReady specification, v2.0. https://github.com/agentready-org/standard/blob/main/spec/spec.md, read 2026-11-04.
- AgentReady repository README. https://github.com/agentready-org/standard/blob/main/README.md, read 2026-11-04.
- AgentReady changelog. https://github.com/agentready-org/standard/blob/main/CHANGELOG.md, read 2026-11-04.
- Ora, AgentReady launch post. https://ora.ai/blog/agentready-launch, read 2026-11-04.
- Ora, check catalog (contract 1.38.0). https://ora.ai/api/checks, read 2026-11-04.
- Cloudflare, isitagentready.com llms.txt. https://isitagentready.com/llms.txt, read 2026-11-04.
- agent-ready.dev, methodology. https://agent-ready.dev/methodology, read 2026-11-04.
- SiteSpeakAI, AI Agent Readiness Scanner. https://sitespeak.ai/tools/ai-agent-readiness-scanner, read 2026-11-04.
- Mintlify, Agent Score. https://mintlify.com/blog/agent-score, read 2026-11-04.
- is-agentic.com, methodology. https://is-agentic.com/methodology, read 2026-11-04.
Last reviewed 2026-11-05