Open anything, thensee what it does.

Guard.ch runs your browsing in a throwaway container in our cloud, so nothing reaches your machine. Beside the page, live analysis shows every request, redirect and script, with a verdict that names its sources. And your AI agent can take the wheel over MCP.

Start an investigation

Open it, watch it, decide

A link, a download, a site you have no reason to trust. Guard.ch opens it in a cloud browser far from your machine, shows you what it does while you use it, and leaves you something to decide on. Or hands the whole loop to your agent.

Open it somewhere else

Right-click a link wherever you meet it, paste it into Guard.ch, or just start a browser and go. Chrome, Brave, Edge, Firefox and more, from the exit location you pick. Entry points:

See what it does

Real Chrome, real tabs, streamed from our cloud. You drive the page like any visitor while the analysis shows redirects, requests, cookies, downloads, tracker domains and technologies, and you pull registry, hosting, certificate and reputation data on the hosts that matter.

What the analysis shows

Decide, or hand it over

Settle the verdict, copy the indicators you need and end the session; nothing of the page ever touched your machine. Or hand the whole loop to your AI agent over MCP and watch it work.

See the whole product

The bank is invented; the pattern is not. Open a link you have no reason to trust, and the lane beside the page tells you what it is doing while you look at it.

Paste a link to openalpinepay-secure.help/loginStart

Autohttps://alpinepay-secure.help/login

AlpinePay

VerdictEvidenceAsk AI

evidence34 events0:00.4alpinepay-secure.help3010:00.6secure.alpinepay-secure.help2000:00.92 technologieslookup0:01.2session_idset0:01.4alpinepay-secure.help/assets/app.js2000:01.63 fingerprint readsfp0:01.9185.22.14.90lookup0:02.1credentialspost0:02.4device profilefp0:02.6fonts.static-cdn.example2000:02.8edge_sessset0:03.2clipboard accessfp

This exact host is a credential harvester wearing a bank's sign-in page.

Registered three days ago behind privacy protection, with no independent source naming it anywhere. The form ships what you type to a different host, and the page reproduces a sign-in screen it has no relationship to. None of that needs a feed hit to be conclusive.

Domain ageRegistered 3 days agoPopularity rankUnrankedCross-references0 independent sources name this hostFeed hitsNone yet: too new for the feedsTLSDV certificate, issued 3 days ago

Key signals

Credentials leave the originThe login form posts to an unrelated host on a different registrable domain.

Reproduces a sign-in page it has no claim toLayout, wording and assets copy the real service; the certificate and the address belong to neither.

Three days old, registrant hiddenRegistry data shows a fresh registration behind privacy protection and no history at all.

Two hops before the page answersThe link passes through a shortener and a redirector before this host responds.

Block this exact host and copy the IOCs for the ticket. Sibling subdomains are judged separately.

Copy IOCsEnd investigation

Suspect sites never touch your network

Your workstation

You work from an ordinary browser tab. The suspect page runs in our cloud and reaches you as pixels and input, so it never touches your endpoint, your corporate network or anything installed on either.

Our egress

Resolution and traffic never originate from your network or your IP. Pick the exit country per investigation, switch to residential egress when a page treats datacenter ranges differently, or route through your workspace's own WireGuard profiles.

A throwaway container

Built when the investigation starts and destroyed when it ends. Cookies, storage and fingerprints never carry over between investigations or between people, so the page has nothing to recognise anyone by.

Give your agent a browser

Point any MCP client at Guard.ch and your assistant gets a real cloud browser of its own: it opens pages, clicks, fills forms, downloads files and takes screenshots, in a session you can watch live and end whenever you want. Nothing runs on your machine.

Your assistantClaude, Cursor, Codex, Gemini, VS Code

Guard.ch MCPOne endpoint, one app key

A cloud browserIsolated, live, yours to watch

A browser of its ownYour agent starts sessions on any of our Chromium browsers, up to three at once, picks the exit location, and closes them when the work is done.
Real hands on the pageNavigate, click, type, fill and submit forms, scroll, drag, upload files, answer dialogs and switch tabs. Or just say in plain language what should happen.
It reads what comes backScreenshots, page text and structured extraction, plus the console, the network and any JavaScript it wants to run itself.
Your own sessions tooHand the agent a browser you started yourself: it drives that session with the same tools and can switch the live analysis lane on, so you see what the page does while it works.
You stay in the roomEvery agent session is listed in your account with a live view. Open it to watch, take the browser over yourself, or end it on the spot.

Set up agent access

A cloud browser for your team, priced per seat

Analyst is one person on one seat. Team starts at three seats, and every seat you add makes each one cheaper. Isolated browsers, live analysis and agent access are in both plans.

Analyst

Everything Guard.ch does, on one seat.

Unlimited investigations, up to three running at once
Eight browsers, each in a throwaway cloud container
Exit locations worldwide, datacenter and residential
Live analysis with AI verdicts and their sources, $5 a day
Agent access over MCP, plus REST and CDP
Extensions for Chrome, Edge, Firefox and Brave
10 GB of residential data per billing cycle
EU data residency and a self-serve DPA

Seats1 seat

One person, one workspace. Teams start at three seats.

$29/ mo

Billed monthly. Cancel any time.

A card starts the trial. Nothing is charged until it ends.

Team

Your whole team in one workspace, on the same browsers.

Everything in Analyst, on every seat
3 to 50 seats, one member per seat, managed by you
Microsoft Entra SSO, tenant auto-join and workspace branding
Up to three investigations running at once for every person
Per seat: 10 GB of data a cycle and $5 of AI a day, pooled
One invoice, and a lower per-seat price as the team grows

Seats5 seats

$111/ mo

$22.20 per seat

Billed monthly. Cancel any time.

A card starts the trial. Nothing is charged until it ends.

Prices are in USD and exclude any VAT or sales tax that applies.

Common questions

What people ask before they put Guard.ch in front of their team. If yours is not here, ask us and you get a straight answer.

Where does our investigation data live, and how long do you keep it?

An investigation lives only while it runs: the nodes that host the isolated browser keep the session in memory and go away with it, and Guard.ch stores no recording of what happened inside. What we keep is account, workspace and billing data, plus session metadata such as the target address and timestamps, in our EU datacenter in Helsinki.

Is this only for suspicious links?

How does capacity work?

Can an AI assistant use Guard.ch?

Yes, that is what the MCP endpoint is for. Point Claude Code, Claude Desktop, Cursor, VS Code, Codex CLI, Gemini CLI, Windsurf or anything else that speaks MCP at Guard.ch with an app key, and it can start browser sessions, browse, fill forms, take screenshots, and read the console and the network. Nothing runs on your machine, every agent session shows up in your account with a live view, and you can take the browser over or end it at any point.

Can our scripts drive Guard.ch?

Can we roll the extensions out centrally, and does SSO work?

The Chrome, Edge, Firefox and Brave extensions can be force-installed through Google Admin, GPO, Intune or a managed policy file, locked to your workspace's SSO slug. Microsoft Entra SSO is configured per workspace with a guided wizard and a live connection test that tells you what is wrong when it fails. Optional tenant auto-join lets sign-ins from a tenant a manager has proven land in the right workspace automatically; it stays off until someone turns it on.

What does the verdict actually mean?

A rules engine sets a floor from concrete signals: hits on authoritative malware and phishing feeds for the exact hostname, independent DNS security filters blocking it, credential posts from fresh domains, downloads that start themselves. The AI read runs on top, can raise severity but never lower it, and names the sources it used. The judgment describes the exact hostname you opened; findings on sibling subdomains are reported separately with their provenance. You see the evidence either way, and you can disagree with the read.

Are you SOC 2 or ISO 27001 certified, and can we get a DPA?

How do we evaluate it?

Everything, free for 30 days.

Nothing is held back during the trial. A card is attached up front and charged nothing until it ends; stopping takes a few clicks.

Start 30-day trial