Put a signed check in front of every action your agents take.
A pilot runs on the same service everyone uses, with the enterprise pieces switched on. Your agents ask CrawlCheck before they read, cite, connect to or transact with a domain; for three weeks they only record the answer. Then you see, in signed numbers, what it would have stopped.
Ask for a pilot a person replies by email; nothing is charged during the 30 days
Who it is for
| Agent platforms | One signed answer before every action, a receipt your users and their auditors can check, and a record of what each site actually served your agent. |
|---|---|
| Hosting companies and registrars | Show each customer what AI agents are told about their site, with the fix steps for your platform, and give their domains an incident switch and a signed business record. |
| Enterprises routing agent actions | Put a checked, logged, explainable decision in front of every action your internal agents take on the open web, with evidence laid out for the AI Act and NIST. |
What the pilot includes
Every line is live today; the links show it.
| Resolve and preflight at volume | 1,000,000 metered lookups a month on the pilot key, x-crawlcheck-lookups-* on every answer, signed decision receipts, your policy as code (strictest wins). |
|---|---|
| Private lookups | Ask by hash prefix so CrawlCheck never learns which site your agent is about to act on; or read the signed daily snapshot and decide fully offline. |
| Single sign-on | OpenID Connect or SAML 2.0 with your identity provider, limited to your email domains. |
| Service level | 99.9% monthly availability for Resolve and preflight, measured from outside Cloudflare, with service credits. |
| Observers inside your networks | Private observers read your own sites from your offices, data centres or cloud regions; their readings stay on your licence. |
| MCP endpoint monitoring | Your MCP servers probed every 15 minutes: downtime, tool-list changes against the version you approved, poisoning, auth findings; a signed verified status for the public ones. |
| Evidence for auditors | Decision receipts laid out against EU AI Act Article 12 and NIST AI RMF, every receipt checked and the pack signed. |
| Incident switch and webhook | Block connect and transact on your own domains in seconds during a compromise; subscribe to incident events for the domains you depend on. |
How it runs
| Week 1 · integrate | Drop in the SDK (npm or PyPI), the MCP preflight middleware or plain HTTPS. SSO and your policy file go in the same week. |
|---|---|
| Weeks 2–3 · shadow | Your agents ask before every read, cite, connect and transact and record the decision without acting on it. Nothing in your product changes. |
| Week 4 · readout | Signed numbers from both sides: decisions by action and verdict, how many you would have blocked and why, latency from your own network, false positives you disputed and how they were resolved. |
What counts as success
- Every agent action in scope asked first, and the answer reached within 300 ms from your network for 95% of calls.
- The decisions you would have blocked hold up when your team reviews them; any you dispute are resolved, signed, inside 48 hours.
- Your auditors accept the evidence pack for the actions in scope.
After the pilot
Stay on the plan that fits from pricing (Network for sites and teams, Growth for API volume), or stop: the pilot key expires on its own and your data can be deleted with a signed record at /manage.
Ask for a licence or a service
Prefer an invoice, or need something not on this page? Leave your details and a person replies; nothing is charged. Running agents at scale? See the 30-day enterprise pilot.