Agent Ready Report documentation

1. Connect

Setup steps for Claude, ChatGPT and other clients are on the product page. From a terminal, with Claude Code:

claude mcp add --transport http agent-ready-report https://scan.agenticplug.ai/mcp

2. Tools

Both tools are annotated read-only, non-destructive and idempotent. scan_website is also marked open-world, because it reads a third-party website.

3. The report

scan_website returns the report twice: as Markdown in the text content, for the assistant to read and quote, and as structuredContent for software. An excerpt of the Markdown for example.com:

# Agent Ready Report: https://example.com
Scanned 2026-09-30 05:10:29 UTC · doesn't appear to sell online

## Summary
- **Agent readiness:** 0 of 16 passing
- **AI search foundations:** 4 of 11 passing
- **Agentic commerce:** not scored, because the site doesn't appear to sell online. …

## Agent readiness: 0 of 16 passing
…
### ❌ robots.txt: Missing or not working
**Finding:** No robots.txt found (HTTP 404).
**What it is:** A small text file at the root of your website that tells automated visitors …
**Why it matters:** It's the first file every crawler reads. …
**Technical:** GET /robots.txt and parse it per RFC 9309.
…
---
Agent Ready Report by agenticplug.ai · https://agenticplug.ai/agent-ready-report
This is an automated, point-in-time diagnostic of publicly available signals. …

The same scan as structured data (trimmed):

{
  "requestedUrl": "https://example.com/",
  "scannedOrigin": "https://example.com",
  "reachable": true,
  "platform": null,
  "commerce": { "detected": false, "signals": [] },
  "botProtection": { "detected": false },
  "tallies": {
    "agentReadiness": { "passed": 0, "scored": 16, "couldntCheck": 0 },
    "commerce": { "passed": 0, "scored": 0, "couldntCheck": 0 },
    "foundations": { "passed": 4, "scored": 11, "couldntCheck": 0 }
  },
  "checks": [
    {
      "id": "robotsTxt",
      "group": "agentReadiness",
      "status": "fail",
      "statusLabel": "Missing or not working",
      "summary": "No robots.txt found (HTTP 404).",
      "explanation": { "title": "robots.txt", "plainEnglish": "…", "whyItMatters": "…", "technical": { … } }
    }
  ],
  "methodologyVersion": "1.4.0"
}

4. How results are scored

Every check gets one status:

• Each group is reported as a tally, such as “4 of 11 passing”: passes out of passes plus fails. “Couldn’t check”, not-scored and informational results are left out of both numbers. There is no overall grade or level.
• Agent-readiness checks are scored for every business, including emerging standards, which are labeled “emerging standard” because missing them is still common.
• Agentic commerce is scored only for sites that sell online. That is detected from storefront markup on the homepage, or from a site already passing a commerce protocol check, and sells_online overrides the detection either way.
• The report is a diagnostic: it explains each finding and never prescribes fixes. The methodologyVersion changes whenever the check logic or wording does.

5. The 35 checks

The id is what reports use and what explain_check accepts. Ask explain_check for any id to get its full explanation, pass criteria and sources.

6. Limits and errors

Problems come back as a tool result with isError: true and a message written for the person asking, never as a protocol error. For example: an address that isn’t a public website, a domain that doesn’t exist, an email address given instead of a website, a site redirecting to a private address, or a rate limit (“please try again in a minute”). A site that blocks or times out the scanner is not an error: the scan completes and those checks read “Couldn’t check”.

7. What the scanner sends

For each scan, the website being scanned receives:

• About two to three dozen requests for public pages and files (the homepage, robots.txt, sitemaps and standard discovery files), identified by the user agent agenticplug-agent-ready-report/1.0 and signed with Web Bot Auth (HTTP Message Signatures), so a website can verify they come from agenticplug.ai. The public key is at https://scan.agenticplug.ai/.well-known/http-message-signatures-directory.
• Four homepage requests using the published user agents of OAI-SearchBot, ChatGPT-User, PerplexityBot and Perplexity-User, to test how the site treats AI crawlers. These are not signed, since they use those crawlers' names, not ours.
• One homepage load in a headless browser run by Cloudflare, to detect WebMCP tools. Cloudflare marks those requests as coming from its browser service.
• Public DNS lookups over DNS-over-HTTPS (Cloudflare, with Google Public DNS as a fallback).
• Never a sign-in, a form submission or a purchase. At most five requests are in flight at once, and fewer once bot protection shows up.

8. Privacy, terms and support

The connector keeps no IP addresses, request headers or identifiers that AI assistants send. Reports are cached for 10 minutes, and anonymous usage statistics (the site scanned, which assistant asked, and the results) are kept for three months. The details are in the privacy policy and the conditions of use in the terms of service. Please keep each report’s attribution line and disclaimer when you show it.

Questions, bugs or a site that you think is scored wrongly: info@agenticplug.ai.