Videos

WATCH NewScan Config

WATCH NewScan Walkthrough

WATCH NewScan MCP

Overview

// free · local · verifies every finding

NewScan is a free, self-hosted scanner that sweeps your whole engagement — APIs, web apps, network & infrastructure, Wi-Fi, and segmentation — and re-runs every hit to confirm it's real before reporting it. A deterministic floor runs with no key; an optional AI layer (BYOK) goes deeper. Everything runs on your box.

→Free to self-host in the browser console — no license, seat fees, or demo call to scan your APIs, web apps and networks locally. API, CLI, MCP and CI/CD automation need NewScan Pro.
→Local-first — target traffic, findings, and AI keys never leave your machine.
→Evidence-backed findings — every finding is reproduced and cross-referenced before it's recorded, with the evidence attached and severity never inflated.
→AI optional, BYOK — fully deterministic with no key; add OpenAI / Anthropic / Gemini / Ollama for deeper reach.
→CI-ready with Pro — SARIF 2.1.0 plus a severity exit-code gate for any pipeline.
→Compliance-calibrated — one score mapped to PCI DSS, SOC 2, ISO 27001, HITRUST, HIPAA, and NIST 800-53.

NewScan findings view — FINDINGS

Confirmed findings ranked by severity, each with reproduced evidence and remediation attached.

NewScan web-app scan console — WEB SCAN

The live web-app scan — crawl, probe, and verify in a real browser, streamed step by step.

NewScan detected technology stack — TECH STACK

The detected technology stack and infrastructure, fingerprinted from responses and cross-referenced to known CVEs.

Use AI where it makes the scan better

// deterministic proof · agentic reach · your choice

NewScan starts with a complete deterministic scan, then uses AI for the work where judgement and iteration add reach. Every AI-led signal is still checked against real evidence before it becomes a finding. Run without AI, bring your own model, use NewScan AI, or let your own agent drive the loop.

MCP / YOUR AGENT · PRO

Give your agent the controls

With NewScan Pro, the local MCP server lets any MCP-capable agent start scoped scans, run checks, read evidence and reports, fix your code, then rescan until the gate passes. Your agent chooses the workflow; NewScan supplies the scanning control plane.

BUILT-IN AGENTS

Go beyond the fixed sweep

Built-in ReACT and campaign agents use the deterministic results, near-misses and shared scan knowledge to plan high-yield follow-ups, craft target-specific probes and investigate attack paths. An adversarial pass tries to refute the result before it is recorded.

AI ON YOUR TERMS

Choose how it runs

Quick and baseline profiles need no model or key. For AI-assisted depth, bring an OpenAI, Anthropic, Gemini, Ollama or compatible-provider key and keep it local, or use integrated NewScan AI with pay-as-you-go billing. The scan engine and evidence bar stay the same.

Want the full agent loop? Install a NewScan Pro license, then connect NewScan to your MCP client →

Scan types

// five surfaces, one tool

API

REST · GraphQL · gRPC · SOAP · MCP · WebSocket

WEB

SPA crawl (Playwright) · XSS · SSTI · CSRF

NETWORK

hosts · ports · TLS · DNS · SMB/SNMP · CVE

WI-FI

rogue APs, device ID & client posture

SEGMENTATION

cross-VLAN reachability · PCI CDE

Findings are shared across modes — a network scan hands discovered web services straight to the API scanner, so nothing falls between tools.

Detections categories

// high level · OWASP API / Web / LLM Top 10

ACCESS CONTROL

BOLA / IDOR · function-level · business-flow abuse

AUTHENTICATION

JWT inspection · weak-secret cracking

INJECTION

SQLi · cmdi · SSTI · XSS · NoSQL · XXE · Prompts

SSRF

server-side request forgery

DATA EXPOSURE

secrets / PII · mass assignment (BOPLA)

DATA EXFILTRATION

out-of-band egress · blind SSRF/XXE callbacks · DNS/webhook exfil

NETWORK & INFRA

TLS / cert · DNS & email · SMB/SNMP · CVE cross-ref

AI / LLM

LLM01 prompt injection · MCP tool poisoning

BROWSER & SESSION TRUST

CSRF · clickjacking · CORS · cookie flags · open redirect · tabnabbing · XS-Leak

BUSINESS LOGIC

race conditions · value tampering · workflow limits

DENIAL OF SERVICE

open resolvers · UDP amplification · resource exhaustion

Plus CRLF injection, file-upload and cloud-storage exposure, deserialization, SAML/OAuth misconfiguration, dependency & CVE cross-referencing, and cross-finding attack-chain correlation. Out-of-band classes (blind SSRF, OOB SQLi/XXE, blind XSS) unlock with NewScan Pro.

Want the exact payloads and checks for your stack? The technology breakdowns go detector-by-detector on GraphQL, gRPC, SOAP, Spring Boot, Kubernetes, JWT/OAuth, MongoDB, Redis, Supabase, Firebase, WordPress, Next.js and AI-backed APIs.

Protocols

// tested natively, not just REST

REST

HTTP / HTTPS APIs

OpenAPI / Postman / HAR import · full injection & auth testing

GRAPHQL

Dedicated GraphQL pack

engine fingerprinting · introspection abuse · batching / DoS · CSRF & injection

gRPC

HTTP/2 + protobuf

reflection discovery · mTLS transport audit · per-field fuzzing

WEBSOCKET

Live channel testing

CSWSH · origin validation · authenticated message-level fuzzing

MCP

AI tool servers

tool poisoning · unauthenticated exposure · indirect prompt injection

SOAP

Auto-detected

classified by content-type / WSDL, tested through the injection engine

Authentication schemes

// scan authenticated, as a real user

BEARER / JWT

bare tokens auto-prefixed; JWTs inspected for weak signing

API KEY

custom header (default X-API-Key) on every request

HTTP BASIC

username : password, Base64 Authorization header

OAUTH2

client-credentials & password grants, token auto-installed

FORM / SPA LOGIN

API login or browser-driven (Playwright) session capture

HMAC SIGNING

per-request HMAC-SHA256/512 with a shared secret

On the roadmap: session cookie · mTLS client certificates · OAuth2 authorization-code + PKCE.

Reports & exports

// one scan, every format your workflow needs

SARIF

SARIF 2.1.0

Machine-readable results for CI code-scanning — GitHub, Azure DevOps, and any SARIF-aware pipeline, with a severity exit-code gate.

MARKDOWN

Markdown report

A human-readable write-up to review or hand to stakeholders — prints or converts to PDF cleanly.

INTERACTIVE TRIAGE

Self-contained HTML

Per-finding status & notes, severity filtering, and one-click Jira / GitHub tickets — triage in the report itself.

HAR

HAR traffic capture

The scan's raw HTTP requests and responses — replay and inspect every exchange in Burp or browser DevTools.

Run it in CI/CD — or from your agent

// NewScan Pro · scan a release before you ship

Local in-band scanning through the browser console is free — including pointing it at an API as the target. API, CLI, MCP and CI/CD automation require an active NewScan Pro license. Install it once in the console (Settings → License) or pass it as NEWSCAN_LICENSE_KEY. An AI provider key is separate: the keyless profiles skip the AI layer, not the license. See NewScan Pro pricing →

AS AN MCP SERVER · PRO

With a NewScan Pro license installed, the running app serves an MCP server on the same port, so any MCP-capable agent can scan, read the findings, fix the code and rescan until the gate passes. Two commands and you're driving it from Claude Code:

$ docker compose up            # the app, console and /mcp on http://localhost:9700
$ claude mcp add --transport http newscan http://127.0.0.1:9700/mcp

> Using the newscan MCP server, scan http://localhost:3000, then fix each
  real finding in this repo and rescan until nothing high or above is left.

Any other client: add an HTTP / streamable-HTTP server at the same URL. Full tool reference and the scan → fix → rescan loop in the MCP docs.

AS A SCRIPT · PRO

With NewScan Pro, NewScan runs headless with a single command and exits with a severity gate code, so any pipeline can scan a build and fail it before deploy — no server, no browser. The keyless quick profile needs no AI provider key; it still needs the Pro license.

$ export NEWSCAN_LICENSE_KEY=…   # your NewScan Pro license
$ python -m newscan https://staging.example.com \
    --scan-mode api --profile quick \
    --fail-on high --sarif out.sarif

Exit 0 clean · 1 a finding met the threshold · 2 the scan couldn't run, including no active license. SARIF uploads straight to GitHub/GitLab code scanning.

GITHUB ACTIONS

Or use the NewScan Action — one step, no install, SARIF into the Security tab. Point it at a preview deploy or an app you start as a job service; make it a required check and a PR can't merge until it passes.

permissions:
  security-events: write   # upload-sarif needs this
  contents: read

steps:
  - uses: NewNormal-Security/newscan-action@v1
    with:
      target: https://staging.example.com
      license: ${{ secrets.NEWSCAN_LICENSE }}
      fail-on: high
  - if: always()
    continue-on-error: true
    uses: github/codeql-action/upload-sarif@v4
    with: { sarif_file: newscan.sarif }

The SARIF upload needs code scanning enabled — free on public repos, GitHub Advanced Security on private ones — so it is marked continue-on-error: a repo without it still passes, and the gate itself already fails the job on findings. Inputs, outputs and three ready workflows (deploy gate, ephemeral service, per-PR preview) are in the action's README. The action and the script above both need NewScan Pro; scanning through the browser console stays free.

Detection	How the callback confirms it
Blind SSRF	The target fetches the canary over HTTP or DNS — proof the server made the request. Separate arms prove an allow-list bypass too (redirect-follow, DNS rebinding, credential and encoding tricks), each with the remediation for how it got through.
SSRF where scanners don’t look	GraphQL arguments found by introspection, image/URL proxies that carry the origin in the path (/unsafe/http://… — Thumbor, imgproxy, /_next/image), and webhook/upstream fetches (OWASP API10). Same canary, surfaces a parameter sweep never reaches.
OAuth / OIDC URI SSRF	The URI params an authorization server is supposed to dereference — request_uri, jwks_uri, sector_identifier_uri, logo_uri — pointed at the listener. A callback proves the server fetches attacker URLs pre-auth.
Prompt-injection SSRF (LLM & MCP)	A canary URL planted in the prompt instructs the model’s browsing/fetch tool to retrieve it; the callback proves the injection actually drove a server-side request. Same arm confirms a poisoned MCP tool reaching out.
Blind XSS	A stored payload fires later — when an admin views it — and loads its script from the listener. Nothing in the response you got ever showed it.
OOB XXE	An external entity resolves against the listener — in a JSON/XML API, a SOAP operation, or an uploaded SVG — confirming the parser is vulnerable when no file content reflects.
Insecure deserialization → RCE	A language gadget (Python pickle, Java URLDNS — JDK-only, no extra library needed) makes the target call out as it deserializes — proving execution before any response is rendered.
CVE RCE callbacks	Log4Shell (CVE-2021-44228), Text4Shell (CVE-2022-42889), Fastjson autotype and Langflow’s unauthenticated /api/v1/validate/code — a lookup reaching the listener proves the version present actually executes, instead of guessing from a version string.
JWT jku / x5u	A forged token points at a JWKS on the listener; the fetch proves the verifier trusts attacker-supplied keys — token forgery and SSRF at once.
Reset poisoning & email header injection	The app’s own mail lands in the hosted SMTP sink — proving it sent mail out, and revealing a host-header-poisoned reset link (account takeover) or an attacker-injected recipient.
Remote include & XSLT fetch	An include/template parameter (RFI) or an XSLT document() pulls the listener’s URL server-side — the recorded pull is the confirmation when nothing renders.
Open DNS resolver & SMTP relay	Network mode: a recursive query for a canary host, and one external→external test message. A DNS callback or a delivery into the sink is the only way to prove the abuse actually works from outside.