It reads what is public
Your DNS records, your website's front page and public indexes. Nothing to install, no passwords, no access to hand over.
What we check
The scanner has 54 working checks in six areas, and the free check runs all but seven. Each one below says what it looks at, why it matters and how far you can trust the result.
Engine 1.57.0, read 25 October 2026
How the scan works
Your DNS records, your website's front page and public indexes. Nothing to install, no passwords, no access to hand over.
A lookup that times out is recorded as unknown. It is never quietly turned into a pass, so a green result is something we saw.
The score travels with a count of the checks that finished. A scan missing a quarter or more of them is marked Incomplete, not given a word.
The checks
Pick an area. Open any check for what it reads and where its limits are.
Email was built on trust: any server can put your domain in the From line. These checks read the DNS records that let a receiving server tell your real email from a fake, and the ones that keep your mail encrypted on the way. This area carries the most weight in the score, because email impersonation is the most common attack and the place most organisations are weakest.
16 checks · 15 in the free check
Without SPF, a receiving server has no list to check a sender against. With a soft ~all it has the list, but only marks other senders as suspect. Two SPF records cancel each other out.
The TXT records at your domain. One v=spf1 record with -all passes, ~all warns, any other all term or none fails, and more than one record fails. The all term counts wherever it sits, and a redirect is followed to find it.
None once the DNS lookup completes.
Scored: it can lower your score.
Each service you add to SPF uses lookups. Go past ten and receivers stop reading your SPF at all, so a record that looks fine on paper protects nothing.
The DNS lookups your SPF record triggers (the limit is 10) and the ones that resolve to nothing (the limit is 2).
The count of empty lookups is a floor, so it never over-reports. With no SPF record there is nothing to count, so it is not graded.
Scored: it can lower your score.
DKIM is the tamper-proof signature on your email. Without it, forwarded mail fails DMARC and your policy can never safely reach reject.
The common DKIM selector names in DNS. On Microsoft 365 the names are known, so a miss is a real fail. Elsewhere a miss is reported as information.
A custom selector name cannot be guessed from outside, so we say so instead of failing you. On the full scan of a domain whose DMARC reports come to us, the selectors seen signing in them are looked up too.
Scored: it can lower your score.
DMARC at none only watches. Until it is set to quarantine or reject, anyone can send email that appears to come from you. It also caps your score: a domain that can be impersonated cannot score above 69. Quarantine cannot score above 89, and nor can reject in test mode, on part of your mail, or with subdomains at none.
The TXT record at _dmarc.yourdomain: the policy, the subdomain policy, how much failing mail the policy covers (pct), and the test-mode tag from the 2026 standard.
None once the DNS lookup completes.
Scored: it can lower your score.
Without MTA-STS, someone on the path can strip the encryption from mail on its way to you, and neither side is told.
The TXT record at _mta-sts.yourdomain.
Presence only, and more than one record counts as none. Whether the policy is switched on is check 12.
Scored: it can lower your score.
If a server cannot deliver to you securely, TLS reporting is how you find out. Without it, those failures are silent.
The TXT record at _smtp._tls.yourdomain.
Presence only, and more than one record counts as none.
Scored: it can lower your score.
BIMI shows your logo in inboxes that support it. It only works once DMARC is enforced, so it is a reward for doing the rest.
The TXT record at default._bimi.yourdomain.
Presence only, and more than one record counts as none. Gmail and Apple Mail also need a mark certificate.
Advice: shown with a recommendation. It never lowers your score.
Every other record here is only as trustworthy as the DNS that serves it. A broken DNSSEC chain can make your whole domain unreachable.
The DS record at your parent zone, and whether a validating resolver confirms a real answer. Signed and valid passes, signed but broken fails, unsigned is information.
Reports unknown when validation cannot complete.
Scored: it can lower your score.
A short key can be cracked, and a cracked key signs fakes that pass every check.
The keys found on the common selectors. RSA under 1,024 bits fails, under 2,048 warns, 2,048 or Ed25519 passes, and an RSA key that cannot be read fails. A key of any other type cannot be sized, so the check is not graded rather than passed.
Only sizes the keys it can see, so a custom selector may not be graded.
Scored: it can lower your score.
The reports are how you learn who is sending as you. A policy with no reporting address is flying blind.
The rua and ruf addresses in your DMARC record, and the permission record an outside destination must publish to receive them.
A receiver can still choose not to send reports, which nothing outside can see.
Scored: it can lower your score.
A record ending +all approves the whole internet, and one ending ?all protects nothing either. An include that points at a lapsed domain can be re-registered by a stranger, who then sends as you with your blessing.
The record's all term, after any redirect, and whether each include and redirect still publishes an SPF record. An all term of +all or ?all fails, and an include or redirect with no SPF record warns.
A dangling include is a risk, not proof that anyone has claimed it.
Scored: it can lower your score.
It is common to publish MTA-STS in testing mode and forget to switch it on.
The policy file itself, fetched from mta-sts.yourdomain. Enforce passes, unless its mx: lines leave out one of your mail servers: then it fails. Testing or none warns, and so does a file that is not a valid policy (no version: STSv1 line, no mode of enforce, testing or none, or a max_age over a year).
Needs the policy host to answer at scan time. We do not test your mail servers' certificates, which senders also check under enforce.
Scored: it can lower your score.
Many receivers mark down or refuse mail from a server whose name and address do not match in both directions.
For each mail server address: the reverse record, and whether that name resolves back to the same address.
Reads your inbound mail servers. A provider that sends from a separate pool is a small blind spot.
Scored: it can lower your score.
With one mail server and no backup, senders hold your mail during an outage, and a long outage makes it bounce.
How many mail servers your MX records name. Two or more passes, even when one provider runs them all.
A resilience readout, not a security failing.
Advice: shown with a recommendation. It never lowers your score.
A listed mail server has its email refused. It is worth knowing, and worth confirming yourself.
The Spamhaus blocklist, for each IPv4 address of your mail servers.
Blocklists refuse questions from cloud networks like ours, so we ask Spamhaus with a dedicated key, on the free check too. When your mail servers cannot be looked up, it says it could not check. We still never let this one change your score.
Advice: shown with a recommendation. It never lowers your score.
Your records say what receivers should do. The reports show what they saw: the senders still failing authentication, a look-alike sending as you, and the signing keys your real mail uses.
The DMARC reports receivers send us for your domain: the messages they counted, the ones that failed authentication and their largest source, and any impersonation incident open against it. Up to ten signing keys seen in them are looked up beside the common ones, so a custom DKIM selector is found. It informs and never changes the score.
Read on the full scan of a domain whose DMARC reports come to us, as a managed domain's do. The free check reads no reports.
Context: background only. It is never graded.
These checks read your website's front page and the encrypted connection to it: the certificate, the versions of TLS it accepts, and the security headers that browsers rely on. They look at the landing page, not pages behind a login. Two look further: your CAA records in DNS, and the encryption on your mail server.
15 checks · 15 in the free check
TLS 1.0 and 1.1 have known weaknesses and browsers have dropped them. A site that still accepts them can be pushed down to them.
Real handshakes with your web server: the versions it accepts, including a direct try of TLS 1.0 and 1.1. It also checks that the cipher it chooses has forward secrecy.
A clean pass is only given when every handshake completed.
Scored: it can lower your score.
An expired or untrusted certificate puts a warning in front of every visitor.
The live certificate chain and its expiry date. Untrusted or expired fails; under 14 days left warns.
We read the certificate your domain itself serves, and the one on www only when your domain serves none.
Scored: it can lower your score.
Without HSTS, a visitor's first request can be intercepted before it is ever encrypted.
The Strict-Transport-Security header your domain sends itself, on its redirect to www as well as its page, and its lifetime, plus where plain http:// sends a visitor and the browsers' preload list. A header that only www sends warns, because browsers apply HSTS to the name that sent it, and so does one that plain http:// skips by redirecting straight to www.
The preload list is a dated copy, so a recent addition or removal can be missed.
Scored: it can lower your score.
A site that answers on plain HTTP hands an attacker on the same network the whole conversation.
How your domain answers a plain http:// request. A redirect straight to HTTPS passes, and so does refusing plain HTTP connections. Anything else, such as a page or a redirect to another http:// address, warns.
Asks your domain itself, not www. It reports unknown if your domain does not point at a server.
Scored: it can lower your score.
A good policy is the strongest defence against injected scripts. A policy full of exceptions is a rulebook nobody follows.
The Content-Security-Policy header, read as current browsers apply it. It flags unsafe-inline, unsafe-eval, wildcard sources (such as * or https:) and a missing default-src. Browsers ignore unsafe-inline beside a nonce, a hash or strict-dynamic, and wildcards beside strict-dynamic in script-src, so neither counts there. A policy that sets script-src, object-src and base-uri needs no default-src.
Grades the shape of the policy on the landing page. It does not prove an attack would work.
Scored: it can lower your score.
A session cookie without Secure, HttpOnly and SameSite can be read or sent where it should not be.
Each cookie the landing page's response sets, and its three flags.
Only sees cookies set before login. Cookies a script sets later are not read.
Scored: it can lower your score.
A version number tells an attacker which known exploits to try.
The Server and X-Powered-By headers, for any software version they name. An out-of-date version of nginx, Apache, PHP or OpenSSL fails, and any other version warns, Microsoft-IIS/10.0 among them. Ubuntu, Debian, SUSE, Amazon Linux and the Red Hat family fix security bugs without changing the version number, so a banner naming one only warns. The banners AWS's load balancer and Azure Storage put on their own responses are not counted: they name the platform's software, not yours.
"Out of date" is as current as our last review of each product's fixed releases, and an end of support the vendor has announced applies on its day. Only nginx, Apache, PHP and OpenSSL can be called out of date. Where the headers name no distribution, a version is judged on its number alone, even if a distribution patched it.
Scored: it can lower your score.
One insecure script on a secure page undoes the security of the whole page.
The landing page's HTML, for scripts, frames, images and stylesheets loaded over http://.
Cannot see what a page adds later with JavaScript.
Scored: it can lower your score.
Without it, an attacker can lay your page invisibly over theirs and borrow your visitor's clicks.
The X-Frame-Options header, or frame-ancestors in your Content-Security-Policy. Either passes when browsers act on it: DENY or SAMEORIGIN, or frame-ancestors limited to 'none', 'self' or named sites.
None.
Scored: it can lower your score.
A browser that guesses can be tricked into running a file as code.
The X-Content-Type-Options: nosniff header.
None.
Scored: it can lower your score.
Web addresses can carry names, references and tokens. This header limits what travels with a click.
The Referrer-Policy header.
Only unsafe-url is flagged, because it sends your full web addresses everywhere. Any other valid policy passes.
Advice: shown with a recommendation. It never lowers your score.
A feature that is switched off cannot be abused by an injected script.
The Permissions-Policy header.
Presence only.
Advice: shown with a recommendation. It never lowers your score.
Without CAA, any certificate authority in the world can issue a certificate in your name.
The CAA records in your DNS. An issue tag passes. No records warn, and so do records with only a reporting contact, or only issuewild, which covers wildcard certificates alone.
None.
Scored: it can lower your score.
A hardened website with a weak mail server is still a weak front door.
Handshakes on three ports: 443 on your website, and 465 and 993 on your main mail server. A port that offers nothing newer than TLS 1.1 fails. So does a weak cipher, one without forward secrecy, or a SHA-1 certificate.
Grades the ports that answer, and a port that is not exposed is not graded. Port 25, where inbound mail arrives, is not tested.
Scored: it can lower your score.
One wrong cross-origin setting lets any website read your pages as the signed-in visitor.
The landing page's cross-origin (CORS) behaviour, its allowed HTTP methods, and whether its Content-Security-Policy, where it has one, includes frame-ancestors.
Landing page only. If the method probe cannot finish, the result says so and grades the rest.
Scored: it can lower your score.
Your domain is the root of your website and your email. These checks look at how it is registered and served: whether it is close to expiring, whether it can be moved without your say-so, and whether forgotten subdomains are open to a stranger.
10 checks · 6 in the free check
Whoever claims the abandoned service controls a page, and a login form, on your own domain.
It reads up to 300 of your hostnames from public certificate logs, the most notable first. For any that point at one of a dozen cloud services we know, it reads the page that service shows. That page is read whatever certificate answers, because an abandoned service has none for your name.
It fails, and caps your score at 59, only where the provider's own page proves anyone can claim the name: today, Amazon S3 and Surge. The rest are warnings to check by hand. They include a target name that no longer exists, Heroku's "No such app" page, and GitHub Pages whether or not the account exists. A hostname that did not answer, or sits on Tumblr or Bitbucket, makes the result unknown rather than clean.
Scored: it can lower your score.
An expired or untrusted certificate on a subdomain puts a warning in front of everyone who uses that service.
The hosts your mail and SPF records name, then the names a stranger tries first, such as vpn. and remote., then your hostnames from public certificate logs, the most notable first, until 30 still resolve, and the certificate each serves on port 443. One that does not validate, has expired or names another host fails; under 14 days left warns.
Port 443 only, on at most 30 subdomains. A certificate another provider serves for its own name, or one reached only through your wildcard DNS record, is noted and not charged. A subdomain that does not answer is named as not read.
Scored: it can lower your score.
A wildcard widens what an attacker can use and masks the typos and leftovers you would otherwise notice.
One lookup of a random name under your domain.
A risk in how the domain is set up, not a proven hole.
Scored: it can lower your score.
A lapsed domain can be bought by anyone, along with every email sent to it.
The registry's expiry and registration dates. Inside 30 days fails and inside 90 warns. A domain the registry shows held for five years or more warns rather than fails in its last 30 days, locked or not. One still unrenewed a week after its expiry date also caps your score at 59.
Reports unknown for the few domain endings whose registry does not publish the date.
Scored: it can lower your score.
If your registrar account is ever broken into, the lock is what stops the domain leaving.
The registry's status flags for your domain.
Needs the registry to publish its records.
Scored: it can lower your score.
When someone finds a problem, this is how they find you before they find a journalist.
The file at /.well-known/security.txt, or the older /security.txt. A page without a Contact, Expires or Email line does not count.
Reported as unknown when a request for the file gets no answer. If your website does not answer at all, it is left out.
Advice: shown with a recommendation. It never lowers your score.
If every name server is on one network and that network fails, your website and email vanish together.
Your name servers and the networks they sit on. Fewer than two warns. So does having them all on one network, unless it is a large managed DNS service we recognise, such as Cloudflare or Amazon Route 53.
Reports unknown when too few name servers can be placed.
Scored: it can lower your score.
If a name server or mail server in your records sits in a domain nobody holds, anyone can register that domain, then answer lookups for your domain or receive mail meant for you.
The name servers and mail servers your domain's own records name. Each is asked for an address, and one with none, outside your domain, has its domain asked of the registry. It fails only where DNS and the registry agree that domain is not registered, and a failure holds the score at 59 until it is fixed. One of the two alone costs points without holding the score.
The free check reads your domain's own name server and mail server records; a full scan also reads the subdomains it finds. A registry that does not answer leaves the finding unconfirmed rather than failed.
Scored: it can lower your score.
Naming habits tell an attacker where to look next. Knowing yours tells you what they can guess.
Past hostnames from public certificate logs.
Description, not a verdict.
Context: background only. It is never graded.
Attackers register domains that look almost like yours, and work out how your staff email addresses are formed. These checks show what a stranger can learn or guess, and whether a look-alike domain is already set up to send email.
4 checks · 4 in the free check
It tells us which hardening advice applies to you.
Microsoft's own public sign-in discovery pages.
Background only.
Context: background only. It is never graded.
A look-alike in a stranger's hands with email switched on is a phishing campaign waiting for a send button. One you registered yourself is good practice.
Up to 30 likely variants of your name, starting with the same name under other endings, such as .com for a .co.uk name. It asks which are registered, which can send email, and who holds them where the registry says.
Most registries hide who holds a name, so many land as "ownership unknown, check". A registered variant is often your own, or another organisation's with the same name. An unusual spelling may not be generated.
Scored: it can lower your score.
A second pass catches look-alikes the first did not think of.
Up to 30 other spellings of your name under the same ending, such as a doubled letter, or a letter swapped for the key beside it. It looks for the ones with mail servers, and an MX record that refuses mail does not count.
Never scored, so the same domain is not counted twice. It does not look up who holds a name, so one with mail servers may be yours or another organisation's. Around a name of three characters or fewer it is information only.
Context: background only. It is never graded.
Once the pattern is known, every name on your website becomes an address to target.
The scan reads no real addresses, so it reports the first.last pattern, marked as assumed.
A best guess, and the same one for every domain.
Context: background only. It is never graded.
Some services publish a searchable index of what every internet address exposes. We read that index for your website's addresses and, on a customer scan, your subdomains'. We also check whether a known shielding network sits in front of your website. Nothing is sent to your systems; we read public records.
5 checks · 3 in the free check
Remote access, databases and admin pages are the first things an attacker looks for.
A public internet index, for the addresses your website is served from: open ports, and the vulnerabilities it lists against them. An address that belongs to a content network or firewall service in front of you is left out. It is recognised by the provider's published addresses or hostnames, or by the index's own record of it.
Open ports are observed. The vulnerability list is the index's inference from version numbers, so it weighs far less and the finding says to confirm it. The scan cannot tell shared hosting from a server of your own. On shared hosting the listing is the provider's to fix, and the finding says so.
Scored: it can lower your score.
Remote access and admin services often sit on subdomains such as vpn. or remote., which the website's own checks never read.
The hosts your mail and SPF records name, then the names a stranger tries first, such as vpn. and remote., then your hostnames from public certificate logs, the most notable first, until 30 still resolve. Up to 20 of their addresses are looked up in a public internet index, for high-risk services and the vulnerabilities it lists against them.
An address that belongs to a content network or firewall service, or that your website already uses, is left out. A load-balanced address the index does not tie to your domain is set aside. Vulnerabilities are the index's inference from version numbers, and on shared hosting the listing is the provider's to fix.
Scored: it can lower your score.
A version number in a banner tells an attacker which published flaws to try, on a host that is often less watched than your website, and a remote-access or mail host is a way into the office network.
The homepage of each subdomain the sweep finds, asked once, as a browser asks. Its Server and X-Powered-By headers are read as the server banner check reads yours: an out-of-date version of nginx, Apache, PHP or OpenSSL fails. One such host costs 15 points and each more 4, up to 30.
Only nginx, Apache, PHP and OpenSSL can be called out of date, as on the server banner check, and a version that is not out of date is named without a charge. No other page is requested and no port is scanned, on at most 30 subdomains. A host that does not answer is named as not read.
Scored: it can lower your score.
A flood aimed at an unshielded server takes the site down with it.
Whether every address your website is served from sits behind a known scrubbing or content network. A site only partly behind one does not count as covered, and the finding names the addresses left exposed.
Spotting a known shield is reliable. One we do not recognise reads as none, and we do not test whether the server behind a shield can be reached directly.
Advice: shown with a recommendation. It never lowers your score.
If you are working towards one of these frameworks, it shows how many of the signals relevant to it are in place.
The scan's own results. Cyber Essentials has two signals: high-risk services or vulnerabilities a public index lists for your website, and out-of-date software named in your server's headers. The ISO standard for information security has three, among them DMARC and clickjacking protection. NIS2 has six, among them DMARC, encryption and the code your home page loads. A signal the scan could not assess is not counted as in place.
Information only, and never scored. It is not an assessment against any of these frameworks: each covers controls an outside scan cannot see.
Context: background only. It is never graded.
Most websites load code from other companies. These checks come from a single read of your home page: old libraries with known holes, code from public library hosts loaded without a tamper check, and the third parties and trackers present.
4 checks · 4 in the free check
A published advisory tells anyone, attackers included, where an old library is weak.
The version written in each script's address: its file name, its folder path, a package@version address, or the ?ver= WordPress puts on a library it ships. Five well-known libraries are covered, jQuery and Bootstrap among them. Each published advisory counts only for the versions it affects.
A short, high-confidence list, read only where an address states a version. Whether a vulnerability can be used against your site depends on how the site uses the library, which this check does not test. On a site that builds its pages in the browser, only the scripts in the page as served are read. If that page loads nothing from outside, it reports unknown.
Scored: it can lower your score.
If a public library host is compromised, an integrity check is what stops an altered file running on your site.
The scripts and stylesheets your home page loads from public library hosts, such as cdnjs, jsDelivr and unpkg, and how many carry an integrity hash.
Home page as served. Other outside code is listed under third-party code and not graded here. That includes tag managers, and your own asset domains under another name.
Scored: it can lower your score.
Whatever each outside host serves is loaded into your visitors' browsers as part of your site.
The hosts outside your own domain that your home page loads scripts and stylesheets from.
An inventory of the page as served, never graded. Code that a script adds once the page loads, such as a tag manager's, is not listed. Some hosts may be your own under another name.
Context: background only. It is never graded.
Each tracker is something your cookie banner and privacy notice must account for.
Known tracking and analytics scripts on your home page.
Listed only when the page as served names one in a script or stylesheet tag. Trackers that a tag manager or a code snippet adds as the page loads are not seen.
Context: background only. It is never graded.
How the score works
Your DMARC policy sets where the score starts: 100 at full reject, 89 when it enforces anything less, and 69 while your domain can still be impersonated. Points come off from there for what the scored checks find, and checks marked advice or context never lower it. Only full reject can reach Excellent.
Some findings hold the score down until they are fixed. A critical one the scan saw for itself, such as an expired certificate on your website or a registration a week past its expiry date, keeps it at 59 or below. A service that should never face the internet, such as remote desktop, keeps it at 89 or below while a public index lists it.
| Score | Word | What it means |
|---|---|---|
| 90 to 100 | Excellent | Little for an attacker to work with. |
| 75 to 89 | Good | Sound, with gaps worth closing. |
| 60 to 74 | Fair | Real weaknesses an attacker would find. |
| 40 to 59 | Weak | Several serious gaps. |
| 0 to 39 | Poor | Open to impersonation and more. |
What this scan is not
Subdomain takeover, hidden hostnames and naming history start from public certificate logs, a shared source with a strict limit. The services, certificates and server software on your subdomains read up to 30 of your other hosts, and DMARC report evidence reads the reports your domain sends us. They run on a customer's own verified domains, not on the free check.
Leaked passwords, infostealer logs, leaked secrets, document metadata and a search for unknown servers have no data source connected. They report nothing, and we do not claim them.
Nothing here tries a password, probes a port or attempts a way in. Anything intrusive needs your written permission and is a separate piece of work.
About 30 seconds. No account. Nothing to install.