Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
nginx-h2-probe — Safe remote scanner for CVE-2026-123007 (nginx HTTP/2 RCE). | Kitploit
Tools/GitHubGitHub/mazeresearch/nginx-h2-probe
Defensive ToolsReconnaissanceVulnerability ScannersWeb Vulnerability ScannersExploitationInformation GatheringWeb SecurityNetwork SecurityPenetration Testing
GitHubmazeresearch/nginx-h2-probe

nginx-h2-probe

1 day agoNot yet reviewed
Share

Safe remote scanner for CVE-2026-123007 (nginx HTTP/2 RCE).

View Repository

nginx-h2-probe

Safe remote scanner for CVE-2026-123007, the 2026 nginx HTTP/2 flow-control RCE. It also flags CVE-2026-123633 (the HTTP/2 use-after-free RCE) from version and config, but does not actively probe it (see Which CVE is actively probed).

The active behavioral test is for CVE-2026-123007. One command tells you whether an endpoint runs the vulnerable code, even when the version banner lies (distro backports keep the old version string) or is hidden. The test can't crash or exploit a server. It reproduces the vulnerability's bookkeeping precondition in a bounded way, not anything memory-unsafe (details below).

Before you run it: it's a single file, so read it first. Only probe hosts you're authorized to test. The active test is safe by design (see Why this is safe to run on production), but it does send a few MB of crafted HTTP/2 traffic.

$ nginx-h2-probe check www.example.com
[*] www.example.com: active test selected -- this sends a few MB of safe,
    non-exploiting test traffic and takes 1-8 minutes.
[.] active attempt 4/14: prefill 128000 pings
[.] active attempt 7/14: prefill 152000 pings
[.] attempt 152000 pings: over-maximum window state reproduced

target: www.example.com:443
ALPN h2 negotiated : True
Server banner      : nginx/1.31.8
CVE-2026-123007 behavior : VULNERABLE (flow-control validation matches unfixed
                            1.9.5-1.31.8 code)
send_window exceeded 2^31-1: YES -- CVE precondition reproduced (bounded, safe)

RESULT: VULNERABLE to CVE-2026-123007 on this endpoint.  (exit code 1)

Run it (no install needed)

Pure Python 3.8+, zero third-party dependencies. The whole tool is the single file nginx-h2-probe.py, so there is nothing to install.

# clone and run (simplest):
git clone https://github.com/MazeResearch/nginx-h2-probe
python3 nginx-h2-probe/nginx-h2-probe.py check www.example.com

# or download just the one file and run it anywhere:
curl -fsSL -O https://raw.githubusercontent.com/MazeResearch/nginx-h2-probe/main/nginx-h2-probe.py
python3 nginx-h2-probe.py check www.example.com

Two optional extras. pip install . from a clone puts a nginx-h2-probe command on your PATH (the examples below use that short form; python3 nginx-h2-probe.py works identically). Docker is handy when you need to run from a host that can reach the origin:

docker build -t nginx-h2-probe .
docker run --rm --network host nginx-h2-probe check www.example.com

Passive-only triage (instant, zero active traffic): add --passive. JSON output for CI: add --json.

Use it with an AI agent

Point a coding agent at this repo and let it run the check. It reads AGENTS.md, which tells it to stay read-only by default, warn you before anything that sends traffic, and only probe hosts you've said you're allowed to test.

Copy-paste prompt:

Clone https://github.com/MazeResearch/nginx-h2-probe and read its AGENTS.md. Check whether HOST is vulnerable to the 2026 nginx HTTP/2 RCE (CVE-2026-123007). Start passive and read-only, and warn me before you send any active traffic or probe anything you're not sure I'm allowed to test.

Which CVE is actively probed

  • CVE-2026-123007 (HTTP/2 flow-control integer underflow, affects nginx 1.9.5-1.31.8): actively probed. The behavioral test returns a VULNERABLE / not-vulnerable verdict based on how the server handles the flow-control sequence, independent of the version banner.
  • CVE-2026-123633 (HTTP/2 connection use-after-free, affects 1.19.7-1.31.8): not probed remotely. Triggering it means exhausting the server's connection slots, which is the DoS/UAF itself. It is reported from the banner version and, with the local subcommand, the reverse-proxy config. If a host is vulnerable to 123007 and reverse-proxies (proxy_pass etc.), treat 123633 as exposed too.

Both fixed in nginx 1.30.7 (stable) and 1.31.9 (mainline).

Why this is safe to run on production

The memory corruption in CVE-2026-123007 only becomes reachable once the connection flow-control window is driven to values near 2^63. This probe drives the window to about 2^31 (the RFC 9113 maximum) plus a small overshoot, well below that range, and uses only ordinary HTTP/2 frames (GET, PING, WINDOW_UPDATE, RST_STREAM). It targets the bookkeeping precondition, not the corruption itself. On a patched server the effect should be ordinary FLOW_CONTROL_ERROR handling, the same as any non-conforming client.

It is still live traffic to someone's server: a few MB per run, 2-8 minutes (worst ~15, up to 14 calibration attempts), and some benign log noise ("http2 flood detected", "client violated connection flow control"). Read the code first, test in a lab if you can, and only run it against hosts you are authorized to test.

What it reports

VerdictMeaningExit
VULNERABLEflow-control validation matches unfixed 1.9.5-1.31.8 code; the bounded over-maximum precondition was reproduced1
not-vulnerable/not-exposedvalidation matches fixed code0
not-exposed (HTTP/2 not enabled)HTTP/2 was not negotiated for the name tested, so the CVEs are not reachable here0
INCONCLUSIVEcould not obtain a decisive answer (notes explain what to change)2
assessment-onlypassive mode: banner/ALPN/version ranges only0

The verdict above is for CVE-2026-123007. CVE-2026-123633 is reported as a version/config assessment only (see Which CVE is actively probed); the local subcommand, run on the nginx host, checks the proxy configuration directly.

Commands

nginx-h2-probe check HOST [options]   # remote check (default command)
    --port 443  --sni SNI  --path /x  --passive  --json  --insecure
    --flood-multiplier 2   # retry hint for INCONCLUSIVE results
nginx-h2-probe local                 # on-the-nginx-host checks
nginx-h2-probe serve [--token S]      # HTTP wrapper (curl-friendly)

A bare target works too: nginx-h2-probe example.com. The package also runs as a module: python -m nginx_h2_probe check example.com.

Library use

from nginx_h2_probe import check

result = check("origin.example.com", active=True)
result["verdict"]             # "VULNERABLE" | "not-vulnerable/not-exposed" | "inconclusive"
result["over_max_confirmed"]  # was the CVE-2026-123007 precondition reproduced?

Scanner service (for systems with nothing but curl)

nginx-h2-probe serve --listen 127.0.0.1 --port 8080 --token SECRET
curl 'http://scanner:8080/probe?token=SECRET&target=origin.example.com&active=1'
# -> {"verdict": "VULNERABLE", "over_max_confirmed": true, ...}

Probes run one at a time. Protect the service with --token, and only point it at targets you are authorized to test.

Limitations

Download Tool

Most Popular

View all →

Discover the most used tools by our community.

  • 1

    OsintHunter

    letox4/osinthunter

    🎯 The ultimate OSINT toolkit: investigation graph, confidence scoring, alias families, map, timeline and reports. 49 modules, 400+ sites, no API keys required.

    1.0k
  • 2

    ghostlock-app

    yukonga/ghostlock-app

    GhostLock One-Tap Execution App (CVE-2026-43499)

    593
  • 3

    looking-glass

    anonghosty/looking-glass

    CSIRT platform that ingests Shadowserver reports into MongoDB and replays them as interactive maps, analysis, findings, advisories and executive dashboards.

    371
  • 4

    redforge

    cybersecurity91/redforge

    Browser-side red team toolkit with 102 offline tools for recon, payload generation, credential attacks, crypto, encoding and engagement reporting.

    336
  • 5

    ARTEX

    baarai/artex

    LLM multi-agent autonomous penetration testing platform with planner/worker agents, asset and exploration graphs, MITM traffic recording, and vulnerability retesting.

    326
  • 6

    CyberStrikeAI

    ed1s0nz/cyberstrikeai

    The system of action for AI-native cybersecurity—where intent becomes governed execution, evidence becomes operational memory, and every operation improves the next.

    322

Explore all tools

Browse our collection of tools

View all tools →