AgentAudit
Your site is invisible to AI agents and you don't know which check is failing. The free scanner tells you what fails; the deep report tells you why, hands you the fix files, and estimates the work.
$29 one report, one domain, paid in USDC — delivered as a static page you own.
What the deep report gives you (vs the free card)
| Free card | Deep report ($29) | |
|---|---|---|
| All five agent-readiness checks | Yes | Yes |
| Evidence appendix (raw curl-reproducible probes) | No | Yes |
| Rendered llms.txt v2 fix file | No | Yes |
| Paste-ready robots.txt AI allow-block | No | Yes |
| Prioritized fix list with effort estimates | No | Yes |
| Machine-readable verdict record embedded | No | Yes |
The free five-check card lives at llmstxtmcp.dev/scanner — start there; buy the deep report when you want the fixes.
Pay like an agent — x402, one call
1.
GET /report/<domain> on the paid rail → a 402 with the standard x402 challenge (USDC on Base, gasless EIP-3009 — you need no ETH for gas).2.Any x402 client signs and retries automatically; the report HTML + settlement header come back in the same call.
3.The self-describing manifest is at
/.well-known/x402.Pay like a human — USDC + one email
USDC on Base:
0x1eAf8042DdF049aEd0eA8D3C1981914a8882BC15
1.Send $29 in USDC (Base) with memo
agentaudit-<yourdomain>.2.Email the transaction hash and your domain to [email protected].
3.You get the report URL back — usually within a few hours, always within a day.
Every report contains
- The five checks — llms.txt (llmstxt.org v2),
.well-known/mcp.json, a reachable MCP HTTP endpoint, docs markdown mirrors /describedbyrelations, npm keyword hygiene. - Evidence appendix — the raw probe output (HTTP codes, byte counts, response excerpts) so every verdict is reproducible with curl.
- A rendered llms.txt v2 fix file — built from what the scan verified, with TODO lines marked; commit it to your site root.
- A paste-ready robots.txt allow-block — 18 AI agent crawlers explicitly allowed, so crawling is never the reason a check fails.
- A prioritized fix list — each fix with why, how, and an effort estimate (S ≤ 2h / M 2-4h / L 1-2d), sorted by priority.
Fair-use notes
The scan runs five probe families against your public pages (plus npm registry lookups) — nothing else. No KYC, no account, no tracking; the report page is delivered by URL and the payment is on-chain. Judgement is automated: the fix list is what the checks imply, not custom consulting.