Check
Free Agent-Ready scan. 12 public checks that tell you whether AI agents can find, read, and buy from your site, with an evidence URL for every finding and a shareable report.
Check
Check is the free Agent-Ready scan at codespar.dev/agent-ready. Enter a domain and it answers the three questions a buying agent asks about your site: can I find it, can I read it while identifying myself as an agent, and can I transact with it. No signup, no install, and nothing is written to your site.
Evidence, not opinion
Every finding carries the URL that was fetched and what it returned. Nothing is inferred or extrapolated. If a probe could not run, the report says so explicitly.
The 12 checks
Each check is a plain HTTP GET against a public, documented convention. Weights sum to 100, so the three category scores add up to the overall score.
| Question | Checks |
|---|---|
| Discover: can an agent find you? | JSON-LD on the homepage (with product-schema detection), sitemap.xml, llms.txt, llms-full.txt, AGENTS.md |
| Trust: can a self-identifying agent read you? | robots.txt, including rules that block AI crawlers by name |
| Transact: can an agent buy from you? | /.well-known/agent-card.json, /.well-known/ucp, /.well-known/oauth-protected-resource, /openapi.json, /.well-known/mcp.json, /.well-known/ai-plugin.json (legacy) |
A 200 OK alone does not pass a check. JSON surfaces must parse as JSON, sitemap.xml must contain a real sitemap, and text files that come back as HTML are treated as a SPA catch-all, not as the file existing.
Dual user-agent
Every URL is fetched first with an honest agent user-agent (CodeSparAgentReady/1.0). If that is refused, Check retries with a browser user-agent so the check still measures what your site publishes. The refusal itself becomes a separate finding on the report: your site is reachable in a browser, but an agent that identifies itself cannot read it. That is exactly what a transparent buying agent experiences.
Checkout-reach probe
On VTEX stores the scan goes one step further and walks the store's public catalog search and order-simulation APIs: catalog search, offer parse, simulation, payment options, and whether Pix is among them. No order is created and nothing is charged.
The report
Every scan produces a public report at /agent-ready/report/<domain>, a link you can share:
- A score out of 100, split into discover, trust, and transact.
- Per-check evidence: the exact URL fetched and what it returned.
- Detected platform (VTEX, Nuvemshop, Shopify, WooCommerce, Magento, or custom) and products read from your site's own structured data.
- Ranked fixes, most severe first. Fix titles are public; the impact detail and the generated starter files (
llms.txt,agent-card.json, a JSON-LD product snippet) come with a free account. - Where your score sits among all scanned domains. Reports refresh hourly.
For store owners: claim and deep probe
You claim a domain by serving a one-line verification token inside the llms.txt or agent-card.json that Check generates for you. Hosting the file proves ownership; no DNS changes needed.
A verified claim is the gate for the deep probe: an owner-initiated run in which a real browser drives your checkout end to end and reports where an agent stalls. Deep probes never run from an anonymous scan, and support VTEX stores in this phase.
From readable to payable
Check is free and stays free. It tells you whether agents can find and read you. When the answer is yes, the next step is letting them pay you:
- Gate puts an x402 paywall in front of any API or MCP server, charging agents per call in USDC. Live on Base mainnet.
- Collect payment links give agents and people one URL to pay you directly, over x402/USDC or Pix. Early access.