Can AI agents actually reach your site?
ChatGPT, Claude and Perplexity send agents to read your site for their users, constantly. If your edge quietly turns them away, you never see it. One scan shows what 14 agents actually get.
14 AGENTS · 2 NETWORKS · 23 CHECKS · NO LLM IN THE SCORING
What the free scan gives you
A straight answer
One verdict: Open Partially blocked Agent-hostile, with the evidence behind it. What can't be measured is labeled, never guessed.
Agent-by-agent detail
Which of the 14 agents got in, which were turned away, and which ones your robots.txt promised to let in. Most owners have never seen that gap.
A score with exact fixes
A 0-100 score from 23 named checks, each with what we saw and a paste-ready fix prompt for your coding agent.
Find out. Prove it. Keep it.
Start with the free scan. Add recorded proof, or weekly monitoring, when you need them.
Free scan
See which of 14 agents can reach your site, what your robots.txt promises them, and exactly what to fix.
Live Agent Test
Real agents attempt three tasks on your site. You get the pass/fail grid across eight disclosed configurations, the exact step where each gave up, cost per attempt, and prioritized fixes. Report in 24 hours.
See a specimen report · Exactly what runs
Coverage and terms
Eight configurations per test, drawn from a roster of 15+ frontier models (Claude, GPT-5, Gemini and more) and 10+ harnesses (Claude Code, Codex and more); at least 6 distinct models and 4 distinct harnesses in every run, and your report names exactly what ran. In your inbox within 24 hours or your money back; if we can't produce a valid report, you're refunded automatically. Checkout, tax and invoicing by Paddle, our Merchant of Record.
Monitor
The full scan re-runs every week. You get the report by email, an alert the moment your verdict or score moves, and an archive of every weekly report.
Billing
$299 per year or $150 per quarter, billed by Paddle, our Merchant of Record. Cancel the renewal anytime; monitoring runs to the end of the paid period.
Need a workflow proven end to end, and fixed?
The Full Workflow Test takes a flow that makes you money (signup, checkout, your API), tests it end to end with real agents, and implements the fixes with your team. Custom scope; we reply within one business day.
RECORDINGS · PRIORITIZED FIX REPORT · FIXES IMPLEMENTED
Real scans, real surprises
From our 12-site validation run (18 Aug 2026). Click one to run a fresh scan; results are measured live when you click.
robots.txt welcomes ClaudeBot, Applebot and Meta-ExternalAgent, but when they actually knocked, the edge turned them away. Invisible from the inside.
robots.txt disallows 11 AI crawlers and the edge enforces exactly that. A declared policy, honestly enforced, earns full credit. We measure honesty, not openness.
Every agent we sent got the page, and the site guides them with llms.txt. This is what agent-ready looks like.
How it works
1 · We knock
Your site, requested exactly as 14 real AI agents request it, from two independent datacenter networks, with a normal-browser baseline.
GET / · UA: ClaudeBot · 200 OK
2 · We compare
What robots.txt declares for each agent, beside what your edge actually did. Then 23 checks measure how readable your site is once an agent gets in.
robots: allow · edge A: 200 · edge B: 200
3 · We judge honestly
A finding counts only when both networks saw the same thing. Anything unmeasured is excluded from the score, never counted against you.
dual-confirm required · unmeasured -> excluded
- NO AI IN THE SCORING
- We knock with the exact request signatures real agents send, but no LLM ever judges the result; every finding is a direct measurement. Want real AI models driving the testing end to end? That's the Live Agent Test.
- WHAT WE NEVER CLAIM
- Our requests come from datacenter networks, where AI agents actually run, but not from the AI vendors' own verified addresses. The strongest claim a report makes: your edge treats agent requests from cloud servers as hostile. Never that a specific vendor's bot is blocked.
- FRESH BY DEFAULT
- Every scan measures your site as it is right now; there is no long-lived result cache. The one exception: an accidental double-submit within 45 seconds can receive the identical just-run result, marked as such, and Re-scan always runs fresh. Completed reports are archived for score history and your report link. Full roster, verdict rules, and all 23 weighted checks: the methodology.
Also an API
The same fresh scan, as JSON. No key; the request carries the same contact details as the form. One scan per domain per minute.
curl -X POST https://agentcapable.dev/scan \
-H 'content-type: application/json' \
-d '{"domain":"yourdomain.com",
"lead":{"name":"...","email":"..."}}'Everything for machines lives at agentcapable.dev/developers: the OpenAPI 3.1 spec, the check catalog, the MCP server, and the npx agentcapable CLI.