agentready

The open-source pre-flight check for agentic commerce. Is your store invisible to AI shopping agents? Find out in 60 seconds.

Live on npm — run it now: npx @mntglobal/agentready your-store.com

AI shopping agents are here — ChatGPT discovers products from merchant feeds, Google's AI Mode transacts over UCP, Shopify exposes MCP endpoints for every store. agentready scans any store URL the way an agent sees it and scores its agent-readiness: a letter grade, category breakdown, and a prioritized fix list with evidence quoted from your actual pages.

npx @mntglobal/agentready https://your-store.com

    B+   85/100 (83.5/98 pts across assessed checks)

  ██████████░░  Structured Data      21.5/25
  ████████████  Agent Access         15/15
  ██████████░░  Product Feeds        10/12
  ██████████░░  Protocol Endpoints   11/13
  ████████████  Machine Readability  10/10
  ██████████░░  AEO Citability       8/10
  ██████░░░░░░  Data Freshness       4/8
  ██████████░░  Accessibility        4/5

  Top fixes:
  1. [SD-04] Add brand, sku, and a GTIN (or mpn) to Product JSON-LD …
  2. [SD-06] Add BreadcrumbList on product pages and Organization …

What it checks — 29 checks, 100 points

| Category | Pts | What agents need | |---|---|---| | Structured Data | 25 | Product/Offer/ProductGroup JSON-LD depth: price, availability, identifiers, ratings | | Agent Access | 15 | robots.txt policy for AI agents (purpose-class scored), llms.txt, no bot-walls | | Product Feeds | 12 | Sitemap product coverage, open catalog endpoints, feed signals | | Protocol Endpoints | 15 | ACP feed-readiness + discovery doc, Google UCP manifest, MCP server detection | | Machine Readability | 10 | Content in initial HTML (agents don't run your JS), clean canonicals | | AEO Citability | 10 | Metadata coherence, FAQ markup, shipping/returns policy discoverability | | Data Freshness | 8 | Sitemap lastmod, HTTP validators, schema-vs-page price consistency | | Accessibility | 5 | Quick pass: alt coverage, lang/labels, heading order |

Full reference: docs/checks.md. Protocol research behind the PE checks (spec versions, sources, probe data): docs/protocol-landscape.md.

Scoring honesty: checks that can't be assessed (unreachable page, no product found) are marked ○ and excluded from the denominator — never counted as failures. Blocking AI training crawlers costs you nothing (that's an IP choice); blocking AI search indexers and live shopping agents is what tanks the grade.

Usage

agentready <url>                          # scan: homepage + auto-discovered product pages
agentready <url> --product <url...>       # assess specific product page(s)
agentready <url> --html report.html       # self-contained shareable report card
agentready <url> --json -                 # machine-readable report to stdout
agentready <url> --ci --min-grade B       # exit 1 below B — wire into CI
agentready <url> --verbose                # evidence + fix for every check

CI gate (GitHub Actions)

- name: Agent-readiness gate
  run: npx @mntglobal/agentready https://staging.your-store.com --ci --min-grade B

Honest limitations

• Static analysis only. We fetch HTML the way most agents do — no JavaScript execution. Client-rendered stores under-score, and the report says exactly what was and wasn't in the initial HTML (that finding is itself the point: agents see the same). A --render mode is on the roadmap.
• Young specs. ACP/UCP/MCP move fast. We pin the spec versions we test against in docs/protocol-landscape.md and re-verify each release. Absence of a public ACP discovery doc is reported as info, never a failure — ChatGPT merchants onboard privately today.
• A scan is ~15 polite requests with a declared UA (agentready/<version> (+https://agentready.mntfuture.com)), timeouts, and no retries. Scan stores you own or have permission to assess.

FAQ

How is this different from a14y? a14y is Timothy Jordan's excellent general agent-readability scorecard (docs/markdown-mirror focus). agentready is the commerce-specific complement: Product/Offer schema depth, catalog feeds, ACP/UCP/MCP endpoint discovery, price/policy freshness. Run both.

My JS-heavy store scores low — is that fair? It reflects what a non-rendering agent experiences. MR-01's evidence shows precisely what was missing from the initial HTML.

Does llms.txt actually matter? No major AI vendor confirms consuming it (Google explicitly ignores it) — we say so in the report and weight it low. But commerce platforms now auto-publish it as agent instructions (Shopify points agents at its shopping skill), which is why it's still worth having.

Development

pnpm install
pnpm test          # fixture-driven, no live network
pnpm lint && pnpm typecheck && pnpm build
pnpm dev -- scan <url>
pnpm record-fixture <url>   # snapshot a real store into fixtures/

Architecture rules in CLAUDE.md · add-a-check recipe in CONTRIBUTING.md · security policy in SECURITY.md.

License

MIT © Magizh NexGen Technologies (MnT)

Built by MnT — we make commerce platforms AI-native and agent-ready. Book a free architecture workshop →