Skip to content

About

Agentless Windows endpoint security posture scanner — audits antivirus, firewall, patching, open ports, services and software, then scores the risk and maps it to NIST CSF 2.0. CLI, desktop dashboard and web API.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

EndpointRadar

EndpointRadar

Agentless Windows endpoint security posture scanner.

Find out how exposed a Windows machine is — antivirus, firewall, patching, open ports, running services and installed software. It scores selected observed issues and explains its findings in plain English.

CI Deploy site License: MIT Python 3.10+ Platform: Windows


PS C:\> endpointradar scan

  WIN-DEMO-01  (10.0.0.51)
  Risk score 9   Level MEDIUM
  Risk level is MEDIUM mainly due to: Risky port listening: 445,
  Potentially risky service running: LanmanServer, LanmanWorkstation.

  Findings (4)
    HIGH     Risky port listening: 445
    HIGH     Potentially risky service running: LanmanServer
    MEDIUM   Potentially risky service running: LanmanWorkstation
    MEDIUM   Unusually low patch count

This historical sample illustrates the output format; stored results and scores are not automatically rescanned or updated when scoring rules change.


Table of contents


Why this exists

Commercial EDR tells you when you are being attacked. Almost nothing tells a small team how easy they would be to attack in the first place.

Basic endpoint hygiene questions are surprisingly hard to answer across even a handful of machines:

  • Is Defender actually running, with real-time protection on — or just installed?
  • Is the firewall off on one profile, which is all an attacker needs?
  • Is SMB (445) listening on a machine that has no business sharing files?
  • What HotFix entries and installed applications are visible for further review?

EndpointRadar collects point-in-time evidence with one command, provides a risk score for selected findings, and explains findings for review. The inventory alone does not prove patch compliance or software vulnerability status.

What it is — and what it is not

It is It is not
A posture and exposure scanner An EDR, an AV, or a real-time monitor
Point-in-time: you run it, it reports Always-on. Nothing is installed or left running
Read-only: it collects and scores A remediation tool. It never changes settings for you
Transparent: scoring is plain Python you can read A model, a heuristic black box, or a cloud service
Offline: no telemetry, no account, no network calls out A SaaS product. Nothing leaves your machine

The repository was previously named KINDA-EDR, which was honest about the gap but said nothing about what the tool does. It has never been an EDR, and the current name reflects what it actually is.

Quick start

git clone https://github.com/4ryanwalia/EndpointRadar
cd EndpointRadar
py -m venv .venv
.venv\Scripts\activate
pip install -e .

endpointradar scan

That is the whole first run. The scan writes scan_results.json into the current directory and prints the summary above.

Try the rest without touching a network:

endpointradar show    -i examples\sample-scan.json      # render the shipped sample
endpointradar threats -i examples\sample-scan.json -v   # NIST CSF analysis with fixes

Installation

The core install has one dependency (psutil). Everything heavier is an optional extra.

pip install -e .                # core: scan, discover, sweep, threats, show
pip install -e ".[remote]"      # + WinRM remote scanning        (pywinrm)
pip install -e ".[report]"      # + Word and Excel reports       (python-docx, openpyxl)
pip install -e ".[gui]"         # + desktop dashboard            (matplotlib, tkinter)
pip install -e ".[web]"         # + web dashboard and JSON API   (Flask)
pip install -e ".[all]"         # everything
pip install -e ".[dev]"         # pytest + ruff, for contributors

You can also run it without installing:

py -m endpointradar scan

The four scan modes

All four produce the same record shape and run through the same scoring code, so their results are directly comparable.

1. Local — the deep, authenticated view

Runs every check against the machine you are on. The richest view available: antivirus state, firewall profiles, visible hotfix entries, listening ports, running services, and installed-software inventory read from the registry.

endpointradar scan
endpointradar scan -f json -o box.json
endpointradar scan -f text                # the long, human-readable report
endpointradar scan --fail-on high         # exit code 2 if the host is HIGH or worse

Run an elevated terminal for complete service and port visibility. Without elevation the scan still works, but some checks report UNKNOWN rather than guessing.

2. Remote — agentless, over WinRM

Stages the stdlib-only subset of the scanner onto the target over WinRM, runs it with the target's own Python, reads the JSON back, and leaves no service or agent behind. The remote host needs no pip packages and no permanent installation.

$env:ENDPOINTRADAR_PASSWORD = 'your-password'
endpointradar remote 10.0.0.51 -u Administrator

# or omit the variable and be prompted
endpointradar remote 10.0.0.51 -u ".\Admin"

Passwords are read from an environment variable or prompted for interactively — never accepted as a command-line argument, where they would land in shell history and the process table.

See Remote scanning over WinRM for target setup.

3. Discovery — who is on this network?

Sweeps a subnet in parallel, resolves DNS and NetBIOS names, reads the ARP table for MAC addresses, and classifies each device as a Windows PC, a mobile device, or unknown, using OS hints, hostname patterns and MAC vendor prefixes. No credentials at all.

endpointradar discover                      # auto-detects your subnet
endpointradar discover 192.168.1.0/24
endpointradar discover --workers 64 --timeout 300
endpointradar discover --include-offline     # list non-responding addresses too
  5 device(s) responding on 192.168.1.0/24
  254 addresses probed.

IP            HOSTNAME MAC               VENDOR  TYPE           STATUS
------------- -------- ----------------- ------- -------------- ------
192.168.1.1   -        2C:97:B1:4E:4B:DF -       Unknown Device Online
192.168.1.83  -        80:64:7C:6B:56:EF Samsung Mobile Device  Online
192.168.1.156 WIN-BOX  -                 -       Windows PC     Online

4. Exposure sweep — the attacker's view

Port-scans every responding device and scores what an unauthenticated attacker on the same network could see. Results are explicitly tagged limited visibility so they are never confused with a full authenticated scan.

endpointradar sweep
endpointradar sweep 192.168.1.0/24 -f json -o exposure.json

Ports checked and their exposure weight:

Port Service Severity Weight Why it matters
445 SMB HIGH 8 Wormable-exploit history and a classic lateral-movement path
3389 RDP MEDIUM 5 Attracts brute-force attempts and remote-access abuse
22 SSH MEDIUM 3 Exposes a remote administration surface
80 HTTP LOW 2 May reveal a management page, app, or banner
443 HTTPS LOW 2 May expose a reachable application or management interface

CLI reference

Nine subcommands. Every one that produces data accepts -f json and -o FILE. -q/--quiet and --no-color work on either side of the subcommand.

endpointradar [-q] [--no-color] [-V] <command> [options]
Command What it does
scan Full posture scan of this machine
discover Find live devices on a subnet, no credentials
sweep Unauthenticated port and exposure scan of a subnet
remote Agentless scan of a remote Windows host over WinRM
threats NIST CSF 2.0 threat analysis derived from a scan
report Word report — fleet summary or per-device threats
show Re-read a saved scan without rescanning
dashboard Launch the Tkinter desktop dashboard
serve Run the read-only web dashboard and JSON API
Full options for every command

endpointradar scan

Option Default Description
-f, --format {summary,json,text} summary text prints the long classic report
-o, --output FILE — Write the result to a file
--no-save off Do not append to scan_results.json
--fail-on {never,low,medium,high,critical} never Exit 2 when the risk level reaches this threshold

endpointradar discover [SUBNET]

Option Default Description
SUBNET auto-detect CIDR, e.g. 192.168.1.0/24
--timeout MS 400 Per-host ping timeout
--workers N 32 Parallel probes
--include-offline off Also list addresses that did not respond
-f, --format {table,json} table Output format

endpointradar sweep [SUBNET]

Option Default Description
SUBNET auto-detect CIDR, e.g. 192.168.1.0/24
--no-save off Do not append to unauth_scan_results.json
-f, --format {table,json} table Output format

endpointradar remote HOST -u USER

Option Default Description
-u, --username required DOMAIN\user or .\localuser
--password-env VAR ENDPOINTRADAR_PASSWORD Env var holding the password; prompts if unset
--timeout S 90 WinRM operation timeout
--no-save off Do not append to scan_results.json
--fail-on LEVEL never Exit 2 at or above this risk level

endpointradar threats

Option Default Description
-i, --input FILE newest saved scan Scan JSON to analyse
-d, --device NAME — Pick a host by hostname or IP
--max N 10 Maximum threats to generate
-v, --verbose off Include descriptions and recommended fixes

endpointradar report [fleet|threat]

Option Default Description
kind fleet fleet = every saved scan; threat = NIST analysis for one device
-i, --input FILE saved scans Source data
-d, --device NAME — Device for the threat report
-o, --output FILE auto Output .docx path

endpointradar show

Option Default Description
-i, --input FILE saved scans Scan JSON to render
-d, --device NAME newest Pick a host by hostname or IP
-a, --all off List every saved scan as a table
--limit N 12 Findings to list
-v, --verbose off Include descriptions and fixes
--fail-on LEVEL never Exit 2 at or above this risk level

endpointradar serve

Option Default Description
--host 127.0.0.1 Bind address
--port 5000 Port
--debug off Flask debug mode

Exit codes

Code Meaning
0 Success
1 Error — bad input, unreachable host, missing file, failed remote scan
2 The --fail-on risk threshold was met or exceeded
130 Interrupted with Ctrl-C

Where files are written

Results land in the directory you run the command from. Override that with an environment variable:

$env:ENDPOINTRADAR_DATA_DIR = "C:\scans"
File Written by
scan_results.json scan, remote
unauth_scan_results.json sweep
report.txt scan -f text (when run through os_check directly)
*.docx report

All of these are in .gitignore — they contain real hostnames, IP addresses and software inventories, and should never be committed.

Using it in CI

--fail-on turns the scanner into a policy gate. It exits 2 once the risk level reaches your threshold, so a Windows runner can fail the build when a machine drifts out of policy.

- name: Endpoint posture gate
  run: |
    pip install -e .
    endpointradar scan --fail-on high --no-color
  shell: pwsh

Machine-readable output for your own tooling:

endpointradar scan -f json -o posture.json
endpointradar threats -i posture.json -f json -o threats.json

This repository's own CI workflow runs the test suite on Linux and Windows across Python 3.10–3.13, then performs a real self-scan of the Windows runner and uploads the result as an artifact.

How risk is scored

Scoring is deterministic and readable — no model, no cloud call, no hidden weighting. The whole thing lives in risk.py and os_check.py.

Overall risk level comes from the total score:

Score Level
0–4 LOW
5–9 MEDIUM
10–15 HIGH
16+ CRITICAL

Component scores are a separate breakdown of observed evidence, with each area clamped to 0–100. They need not add up to the overall score. A zero score means no issue was scored in that area, not that the control is verified or the device is secure.

Component Evidence considered
Antivirus Confirmed not found, disabled, or real-time protection off; collection errors are unknown
Firewall Profiles explicitly reported OFF
Patching HotFix inventory is partial; no points are assigned based on its count or age
SMB / Open Ports Selected listening or reachable ports, depending on scan mode; listening does not prove network reachability
Services Highlighted running services
Third-party Software Inventory only; software versions are not checked against advisories and add no score
Device Protection Secure Boot or system-drive BitLocker protection explicitly disabled; 2 points each

Verify update compliance in Windows Update or your management console. A small or old HotFix list is not proof of missing updates. Software inventory and install dates do not establish vulnerability or support status; check vendor advisories. Older saved scans can contain findings generated by earlier heuristics and should be rescanned before relying on their results.

Unauthenticated (limited-visibility) records are scored only on port evidence. A credential-free result will never claim a machine has no antivirus — it reports that it could not tell.

Authenticated scans report Microsoft Defender tamper protection separately as true, false, or null (unknown) under antivirus.defender_tamper_protection. Older scans may omit this field. It describes Defender only, does not establish the protection settings of other antivirus products, and does not affect the risk score. A missing property, unsupported cmdlet, or query failure is unknown.

Authenticated scans also read the Windows UEFICA2023Status registry value as secure_boot_servicing.state: updated, not_started, in_progress, or unknown. This is reported separately from whether Secure Boot is enabled, and does not change the risk score. Missing, unavailable, or unrecognized registry values are unknown; older scans may omit the field. The result reports the Windows servicing status and does not establish complete firmware or device security. No registry setting or certificate is changed by this check. See Microsoft's status definitions.

Threat analysis and NIST CSF 2.0

endpointradar threats interprets an existing scan — it never re-runs checks or modifies the record. Each finding becomes a threat with a likelihood (1–5) and an impact (1–5); their product is the threat score.

Score Threat risk level
1–5 LOW
6–14 MEDIUM
15–25 HIGH

Categories map to NIST CSF 2.0 functions:

Category CSF function
Firewall, Open Ports, OS / Patching, Services, Third-party Software Protect
Antivirus (detection capability disabled or missing) Detect
endpointradar threats -v

Threats are derived from selected evidence, not a guarantee of exploitability. HotFix age/count and software inventory alone do not create a patch or software-exploitation threat.

Scan record format

Every mode emits the same JSON shape. A full anonymised example ships at examples/sample-scan.json.

Operating-system version and build are inventory only. The scanner does not establish Windows support lifecycle status or update compliance; verify these separately for the device's edition and servicing channel. A missing version is not scored as a failed baseline.

{
  "hostname": "WIN-DEMO-01",
  "device_id": "10.0.0.51",
  "target_ip": "10.0.0.51",
  "risk_score": 9,
  "risk_level": "MEDIUM",              // LOW | MEDIUM | HIGH | CRITICAL
  "connection_status": "Local Scan",   // Local Scan | Online | Unauthenticated ...
  "last_scan_at": "2026-09-14T10:15:00+00:00",

  "vulnerabilities": [
    {
      "title": "Risky port listening: 445",
      "severity": "HIGH",              // LOW | MEDIUM | HIGH
      "description": "Plain-English detail, what it means, and why it matters.",
      "fix": "The concrete remediation step."
    }
  ],

  "system_info":  { "hostname": "...", "ip_address": "...", "os_name": "...", "os_version": "..." },
  "antivirus":    { "found": true, "status": "ENABLED", "realtime_protection": true, "defender_tamper_protection": true, "products": [] },
  "patches":      { "total_patches": 12, "days_since_last_patch": 9, "patch_warning": "" },
  "firewall":     { "profiles": { "Domain": {"state": "ON"}, "Private": {}, "Public": {} } },
  "open_ports":   { "listening_ports": [], "risky_open": [{ "port": 445, "severity": "HIGH" }] },
  "services":     { "running": [], "highlighted": [{ "name": "LanmanServer", "severity": "HIGH" }] },

  "installed_apps": [{ "name": "...", "version": "...", "publisher": "...", "install_date": "..." }],
  "risky_apps":     [{ "name": "...", "version": "...", "risk_level": "HIGH", "reason": "..." }],

  "device_type": "Windows",            // Windows | Mobile | Unknown
  "device_type_label": "Windows PC",
  "scan_supported": true
}

Desktop dashboard

pip install -e ".[gui]"
endpointradar dashboard

A dark-themed Tkinter dashboard over the same engine: a device list, risk gauges and matplotlib charts, per-device drill-down across findings, external exposure, software inventory and live processes, plus one-click Word and Excel export. Discovery, unauthenticated sweeps and WinRM scans all run from here with live progress and ETA.

Web dashboard and JSON API

pip install -e ".[web]"
endpointradar serve --port 5000

A read-only Flask view over scan_results.json, useful for presenting results on a second screen or wiring into something else.

The dashboard includes searchable, severity-filtered findings and CSV export of the current view. Open it through the local server at http://127.0.0.1:5000/, rather than opening the HTML file directly. The API does not grant cross-origin access; scan data can contain sensitive device details. Keep the server bound to localhost unless you have independently secured access to it.

Route Returns
GET / Rendered HTML dashboard
GET /devices Every saved device with summary risk data
GET /device/<device_id> Full detail, component cards and threat analysis
GET /device/<device_id>/live-applications Running-process list for that device

It binds to 127.0.0.1 by default. It has no authentication — do not expose it to a network you do not control.

Word and Excel reports

pip install -e ".[report]"

endpointradar report fleet  -o fleet-audit.docx      # every saved scan
endpointradar report threat -d WIN-DEMO-01 -o box.docx   # NIST analysis for one device

The fleet report covers scope, methodology, per-device findings and a risk summary. The threat report expands one device's NIST CSF analysis with likelihood, impact and recommendations. Excel export is available from the desktop dashboard.

Remote scanning over WinRM

On the target machine, as administrator:

Enable-PSRemoting -Force
Set-Item WSMan:\localhost\Client\TrustedHosts -Value "<scanner-ip>" -Force

The target also needs Python 3.10+ on PATH (the runner locates it via the py -3 launcher).

Then from the scanner:

pip install -e ".[remote]"
$env:ENDPOINTRADAR_PASSWORD = 'your-password'
endpointradar remote 10.0.0.51 -u Administrator

What actually happens: EndpointRadar creates C:\temp\endpointradar_remote_scan on the target, base64-streams the five stdlib-only modules into it, runs a small throwaway runner script, reads back the resulting JSON, and reports. Authentication uses NTLM. Nothing is installed, no service is registered, and no scheduled task is created.

Security limitation: This is a predictable, shared staging directory; the staged Python code and sensitive scan JSON persist on the target after a scan. Concurrent scans of the same target can overwrite each other's files. Anyone able to modify the directory may be able to influence executed code or results. Use remote scans only when the target and directory are trusted and access-controlled, then securely remove the staging directory when no scan is using it.

The current client connects over WinRM HTTP on port 5985 and does not offer an HTTPS option. NTLM is used; Basic authentication is not attempted. HTTP is not equivalent to TLS. Use this mode only on a trusted network, and follow your organization's WinRM security policy.

Requirements

Scanning target Windows 10 / 11 / Server. The checks call WMI, PowerShell, netsh and the registry
Python 3.10 or newer
Core dependency psutil
Privileges Works unelevated; an elevated terminal gives full service and port visibility
Analysis on any OS show, threats, report and the test suite run fine on Linux and macOS against saved scan JSON

Project structure

EndpointRadar/
├── src/endpointradar/
│   ├── cli.py              # the command line interface
│   ├── os_check.py         # local Windows checks (stdlib + psutil only)
│   ├── installed_apps.py   # registry software inventory (stdlib only)
│   ├── risk.py             # component scoring and explanations (stdlib only)
│   ├── scanner.py          # orchestration, discovery, sweep, WinRM staging
│   ├── threats.py          # NIST CSF 2.0 threat analysis
│   ├── reporting.py        # Word report generation
│   ├── gui.py              # Tkinter desktop dashboard
│   ├── server.py           # Flask read-only API
│   └── templates/
├── docs/index.html         # the GitHub Pages site
├── examples/sample-scan.json
├── tests/
└── .github/workflows/      # ci.yml, pages.yml

The four modules marked stdlib only are exactly the set staged onto remote hosts — that constraint is why they carry no third-party imports.

Development

pip install -e ".[all,dev]"
pytest -q
ruff check src tests

The test suite runs from the checked-in sample scan rather than touching the host, so it passes on every platform and exercises the same scoring code a real Windows scan uses.

To preview the site locally:

cp examples/sample-scan.json docs/sample-scan.json
py -m http.server 8765 --directory docs

Security, privacy and legal

  • Scan only what you are allowed to scan. discover and sweep send traffic to every address in the range you give them. Use them on networks you own or have written permission to test.
  • Nothing leaves your machine. No telemetry, no accounts, no outbound calls. Results are local JSON files.
  • Scan output is sensitive. It contains hostnames, IP addresses, MAC addresses and a full software inventory. Every output file is gitignored by default — keep it that way.
  • Credentials are never arguments. WinRM passwords come from an environment variable or an interactive prompt, so they stay out of shell history and the process list.
  • Findings need human judgement. Some checks (third-party AV state, patch visibility) can report UNKNOWN on a locked-down host. The tool says so rather than assuming the worst.

Found a security issue? See SECURITY.md.

Contributing

Issues and pull requests are welcome — especially new checks, better vendor/OUI coverage, and non-English remediation text. See CONTRIBUTING.md.

Good first contributions:

  • Add a port to the exposure map in scanner.py
  • Add a risky-service signature in os_check.py
  • Add a maintained, evidence-backed software advisory source
  • Add a MAC OUI prefix to KNOWN_OUI_VENDORS

License

MIT © Aryan Walia


If EndpointRadar saved you some time, a ⭐ helps other people find it.

Live demo · BugSnaps report workspace · Report a bug · Star the repo

About

Agentless Windows endpoint security posture scanner — audits antivirus, firewall, patching, open ports, services and software, then scores the risk and maps it to NIST CSF 2.0. CLI, desktop dashboard and web API.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages