Full product guide

Overview

Vulnscape checks websites, apps, CRMs and code — especially ones built with AI tools — for vulnerabilities and mistakes, big or small. Deterministic scanners gather evidence; Vulnscape AI (powered by Claude) turns that evidence into severity-scored findings with copy-paste remediation and an aggregate risk score (0–100).

Typical flow:

1. Create or pick a target (URL, host, code, manifest, passwords or share link).
2. Run a scan from /app/scan/new.
3. Open the scan detail page — findings, evidence, contacts, chat, share, compare, export.
4. Optionally mark the target as monitored and re-check it on demand from the Alerts page.

The product console is at /app. Pricing, marketing and this documentation are public.

Getting started

1. Pick a plan on the pricing section and complete Stripe Checkout (or contact us if online checkout is unavailable).
2. Vulnscape creates your account and emails a temporary password.
3. Sign in at /login. Use “Forgot password” to reset at any time.
4. Open New scan and choose a scanner tab.
5. Review findings on the scan detail page; export a report or ask Vulnscape AI what to fix first.

Console layout

After sign-in, the left navigation covers:

• Dashboard — portfolio risk, severity counts, recent scans, highest-risk targets, onboarding checklist.
• Scans — history of every run; jump to New scan.
• Leads Agency — CRM + on-demand lead finder.
• Targets — saved assets, monitoring toggles, latest risk.
• Alerts — new findings, risk increases and failed re-scans, plus the “Check for new alerts” button.
• Reports — export completed scans (Markdown / HTML → PDF).
• Team — organisation + invite seats.
• Integrations — status of email, Slack, Stripe and Sentry.
• Settings — plan, analysis engine status, notification preferences.

A global Vulnscape AI assistant bar sits in the console footer for portfolio-level questions. Each scan also has its own chat panel.

Plans & access

• Ops ($299/mo or $2,999/yr) — Code, Website, Dependency, Password, Network and Secret-leak scanners; Vulnscape AI; on-demand monitoring; alerts; reports; share links; assistant; team basics.
• Agency ($1,299/mo or $12,999/yr) — everything in Ops plus the Leads CRM, the on-demand job-board lead finder, saving scan contacts as leads, and team seats.

Admins can assign or override plans and create time-limited vouch accounts from /app/admin/users.

Code scanner

Where: New scan → Code.

Paste source or upload a file. A regex pre-pass flags SQL injection sinks, XSS sinks, hard-coded secrets/credentials, weak crypto and dangerous functions. Vulnscape AI then reviews the code in context and reports real issues with remediation.

Use it for: your own repositories or client code you are contracted to audit — never for private code you are not authorized to review.

Website scanner

Where: New scan → Website. Requires the I am authorized to scan this website checkbox.

From one seed URL Vulnscape:

• Deep-crawls same-origin links (SSRF-guarded; private/localhost blocked by default).
• Checks security headers, cookie flags, CSRF posture on forms, TLS, mixed content.
• Reviews HTML / forms and file-upload accept= gaps.
• Discovers security contacts (not treated as a vulnerability — information only).

With deep active tests enabled (default):

• XSS & SQLi probes — GET/POST indicators on the most interesting crawled pages (no file uploads to the target).
• Nuclei — known-vulnerability / misconfiguration templates via the tools sidecar.
• SQLMap — capped risk/level confirmation against injectable-looking URLs and forms.
• Heartbleed (CVE-2014-0160) — TLS heartbeat probe on HTTPS.
• Shellshock (CVE-2014-6271+) — non-destructive CGI RCE probe against discovered and common CGI paths.
• Default-admin / lockout probe — capped dictionary + brand-derived credentials against login surfaces. Successful weak login = CRITICAL; no lockout after many failures = HIGH brute-force exposure.
• HTTP rate-limit / abuse protection probe — a capped burst of GETs looking for HTTP 429, Retry-After, RateLimit-* headers and common WAF throttle cues. HIGH when none is observed under the burst (a DDoS or abuse spike can keep being served and inflate hosting bills); INFO when limiting is confirmed. This is not the New scan “Rate limit requests” checkbox, which only paces Vulnscape’s own outbound traffic.

Long website scans run in the background. The scan detail page polls until the status leaves RUNNING. Raw evidence is stored even when zero findings are produced.

Secret-leak scanner

Where: New scan → Secrets.

Paste a public share URL — Claude Code / ChatGPT share pages, GitHub gists, Pastebin or any public http(s) paste. Vulnscape fetches the content (SSRF-guarded), extracts text and hunts for accidentally leaked credentials so owners can be warned to rotate.

What it detects:

• API keys & tokens — Anthropic, OpenAI, AWS access key IDs, GitHub (classic + fine-grained PAT), GitLab, Slack, Stripe live keys, Google, SendGrid, Twilio, npm, Hugging Face, JWTs.
• PEM private keys.
• Crypto material — Ethereum private keys (context-gated), BIP39-style seed phrases.
• Database connection strings with embedded credentials.

Live-validate (optional, read-only):when checked, Vulnscape calls each provider's documented read-only endpoint (e.g. OpenAI GET /v1/models, GitHub rate-limit, Stripe balance) to see if the key still authenticates. Active keys are CRITICAL. Validation never mutates provider state and never moves crypto funds. Cap: 15 checks per scan.

Redaction: raw secrets are masked (sk-ant-…wxyz style) before storage, before Vulnscape AI sees the evidence, and in reports. Use this to warn owners — Vulnscape does not auto-notify third parties.

Dependency scanner

Where: New scan → Dependencies.

Paste or upload package.json, a lockfile, or requirements.txt. Pinned versions are matched against the OSV.dev vulnerability database. Each CVE is explained with an upgrade path.

Password audit

Where: New scan → Passwords.

Enter up to 20 passwords (one per line). Each is scored locally for length, entropy, character classes, keyboard sequences and common-password lists, then checked against Have I Been Pwned using k-anonymity (only a SHA-1 hash prefix is sent). Raw passwords are never stored, and Vulnscape AI receives metrics only.

Network scanner

Where: New scan → Network. Requires authorization.

Probes common TCP ports, resolves DNS, evaluates SPF/DMARC email posture and performs deep TLS certificate inspection — including Heartbleed on port 443 when applicable. SSRF-guarded: localhost, private and link-local ranges are refused.

AI/GPU infrastructure scanner

Where: New scan → AI/GPU Infra. Requires authorization.

Fingerprints a host for exposed, unauthenticated AI/ML services and control planes — common on GPU boxes and pods — and reports each as a severity-ranked finding: Ray dashboards (CVE-2023-48022 “ShadowRay” → RCE), Jupyter, Ollama, ComfyUI, vLLM/TGI, Triton, MLflow, TensorBoard, Gradio, unauthenticated Redis, and open Docker Engine / Kubernetes / kubelet APIs.

Every check is a non-destructive, read-only fingerprint — it confirms a service is present and unauthenticated via a version/health endpoint, and never submits a job, invokes a model or reads data. SSRF-guarded and authorization-gated, like the Network scanner; pass a custom port list under advanced options or use the AI-infra default set.

Advanced scan options

Available on New scan for any scan that hits a remote target (Website, Network, AI/GPU Infra, Secrets):

• Custom headers — one Name: Value per line (e.g. Authorization: Bearer … for authenticated scans). Applied to the crawler, probes, auth check, Nuclei and SQLMap.
• Rate limit — paces outbound requests (~400 ms) and lowers tool concurrency to avoid tripping protections on sensitive systems.
• Proxies — one proxy URL per line, rotated round-robin. http/https for the crawler; the Nuclei/SQLMap sidecar also accepts socks.

You can also tick Add to monitored targetswhen creating most scan types (not Password) — see Monitoring & alerts below.

Vulnscape AI

Vulnscape AIis Vulnscape's analysis engine (Claude). It converts scanner evidence into structured findings — title, severity (CRITICAL → INFO), category, description, evidence snippet, remediation and references — and computes a consistent risk score from severities.

• Per-scan chat— ask what to fix first, how to remediate, or what a score means. Grounded only in that scan's findings.
• Portfolio assistant — console footer chat across your recent portfolio context.
• It does not invent CVEs. If evidence is clean, findings stay empty and the summary says so.
• For secret-leak scans, only redacted evidence is sent to the model — never raw keys.
• The model only runs when you start a scan, re-scan, lead search or chat. There is no background usage.

Scan detail & compare

Open any scan from Scans or the Dashboard. While the status is queued or running, the page polls automatically. When done you get:

• Executive summary, risk score, severity breakdown.
• Each finding with evidence and remediation.
• Stored scan evidence / telemetry (crawl map, tool notes) when available.
• Contact panel (from website scans) — save contacts as leads on Agency.
• Export, share, compare, delete and per-scan chat.

Compare diffs findings and risk against a prior (or chosen sibling) scan on the same target — useful after remediation or a re-scan.

Share links

From scan detail, create a password-protected share link at /share/[token]. Recipients unlock it with the password you send them. Links can expire and be revoked. Shared pages are not indexed. Use this for client deliverables without giving console access.

Leads CRM & lead finderAgency

Where: /app/leads.

Pipeline for inbound requests and outbound prospects. Statuses: New → Contacted → Qualified → Won / Lost. Add leads manually or capture them from a scan's contact panel.

Find leadspulls companies actively hiring engineers from public job APIs (HN Who's Hiring, Remotive, RemoteOK, Arbeitnow, The Muse, Jobicy, optionally Adzuna). With “Qualify with Vulnscape AI” ticked, Claude drops recruiters/staffing agencies, writes a one-line rationale and may infer a website (marked unverified when guessed). Results de-duplicate against your CRM. It runs only when you click the button. We do not scrape sites that forbid it (e.g. Indeed / LinkedIn).

Targets

/app/targets is your asset inventory — name, type, reference, latest risk, scan count and monitoring controls. Re-scan or delete from here. New scans create a target automatically when you supply a name.

Monitoring & alerts

Turn monitoring on for a target (at creation or from Targets). Nothing runs on a schedule: click Check for new alerts on the Alerts page to re-scan every monitored target, or Re-scan now on a single target. Re-scans run in the background — deep website scans can take several minutes each — and the Alerts page shows progress.

Each re-scan is diffed against the previous completed scan. New findings, risk increases and failed re-scans raise alerts in /app/alerts (unread badges in the sidebar) plus optional email / Slack notifications configured in Settings.

Reports & export

From /app/reports or any completed scan detail, export Markdown or a print-ready HTML view (browser Print → Save as PDF). Reports include the summary, risk score, severity breakdown and each finding with evidence and remediation — ready for client deliverables.

Team

/app/team — create an organisation, invite seats via invite link, and list members. Operator accounts are typically provisioned by an admin with a plan assignment.

Integrations

/app/integrations shows configuration status for Resend (email), Slack webhooks, Stripe billing and optional Sentry. OSV.dev CVE matching, Have I Been Pwned and Claude (Vulnscape AI) are always part of the core platform when keys are configured server-side.

Settings

/app/settings — view your plan, the analysis-engine status (API key, workspace, model) and the number of monitored targets; configure email and Slack notification preferences (severity floor).

Security & safeguards

• Authorized use only. Website and network scans require an explicit authorization confirmation.
• SSRF guards. Localhost, private and link-local ranges (including cloud metadata 169.254.169.254) are blocked. Secret-leak fetches re-check every redirect hop.
• No file uploads to targets. Probes never upload files; upload checks are accept-policy / reflection based.
• Secret redaction. Leaked keys are masked before persistence and before AI analysis. Validation is read-only and capped.
• Login lockout. Console sign-in locks after repeated failures (per email and per IP).
• API rate limits. Short-burst and sustained per-IP ceilings; tighter budgets on scans, AI and lead searches.
• Hardened headers. Strict CSP, HSTS and anti-clickjacking on every response.

Admin

Admin-only console pages:

• /app/admin/users — create users, assign Ops / Agency, rename display name/username, create time-limited vouch accounts, extend expiry.
• /app/admin/leads — inbound requests from the public audit/quote form (New / Contacted / Closed).

FAQ

Does the website scan check rate limits?

Yes, in active mode. A capped GET burst looks for 429 / Retry-After / RateLimit headers and WAF cues. Missing throttling under that burst is reported as HIGH (DDoS/abuse → hosting-bill risk). The “Rate limit requests” checkbox only slows Vulnscape’s own scanner — it is not the same check.

Does the secret scanner notify key owners?

No. It detects, redacts, optionally validates read-only, and reports so you can warn owners responsibly. It does not email strangers or move funds.

Why is my website scan still running?

Deep crawls with Nuclei and SQLMap can take several minutes. The detail page polls until the scan is done or failed. A wall-clock safety timeout stops hung jobs from staying in the running state.

Will Vulnscape AI invent findings?

No. Findings must be supported by scanner evidence. Empty evidence → empty findings with a clean summary.

Does anything run automatically?

No. Scans, re-scans and lead searches only start when you click. Monitoring simply marks which targets are included in “Check for new alerts”.

Is this for unauthorized hacking?

No. Vulnscape is for authorized testing and your own assets. Misuse is prohibited.