Domain Risk Signals — Agent Safety Check

Give it a domain or a URL, get back structured trust signals: how old the registration is, whether DNS and mail are set up like a real operation, whether TLS actually validates, and what the live HTTP response does.

Built for AI agents that hit a link and need to decide before fetching it. One call per domain, plain JSON out, no account or API key on the target side.

What it answers

• How old is this domain? Registration date straight from the registry over RDAP. A domain created nine days ago is a different proposition to one created in 2009.
• Is it set up like a real business? MX records, SPF, nameservers, resolvable A/AAAA.
• Does TLS actually validate? Issuer, validity window, and whether the chain is trusted — not just "is there a padlock".
• Where does it really send you? Full redirect chain, final URL, and a flag when it leaves the host you asked about.
• Is the registry unhappy with it? clientHold, serverHold, pendingDelete and similar states.

Output

One dataset record per domain:

{
  "domain": "example.com",
  "riskScore": 15,
  "riskLevel": "low",
  "signals": [
    { "code": "no_mail_records", "detail": "no MX record on the registrable domain", "weight": 10 },
    { "code": "no_spf", "detail": "no SPF record", "weight": 5 }
  ],
  "domainAgeDays": 10847,
  "registeredAt": "1995-08-14T04:00:00.000Z",
  "registrar": "RESERVED-Internet Assigned Numbers Authority",
  "resolves": true,
  "hasMx": false,
  "tlsValid": true,
  "tlsIssuer": "DigiCert Inc",
  "httpStatus": 200,
  "httpFinalUrl": "https://example.com/"
}

riskScore runs 0–100 and is additive and fully explained: every point that lands on the score appears in signals with its own weight. Nothing is hidden, so an agent can apply its own threshold or ignore the score entirely and reason from the raw signals.

riskLevel is a convenience bucket: low under 30, medium 30–59, high 60+.

Input

Field	Type	Default	Meaning
domains	array of strings	—	Domains or full URLs. URLs are reduced to their hostname.
checkHttp	boolean	true	Set false for a passive check with no request to the target.
timeoutMs	integer	8000	Per-lookup timeout.

What this is not

It is not a malware scanner and not a blocklist. It reports observable facts about how a domain is set up, plus a transparent score built from them. A brand-new domain with no mail records and a week-old certificate is not proof of anything — it is a reason for an agent to be careful. Treat the output as evidence, not a verdict.

Notes

• Registration data comes from RDAP, the registries' own structured protocol. Some ccTLDs publish little or nothing; when that happens you get registration_unknown rather than a guess.
• With checkHttp: false nothing is sent to the target's web server at all.
• Domains are checked concurrently, so a list of a few hundred finishes quickly.