ssrf-probe

An authorized SSRF (Server-Side Request Forgery) testing toolkit: it generates a curated set of SSRF probe payloads, injects them into a target request, and tells you which ones are worth verifying — using response analysis and, for blind SSRF, an out-of-band (OAST) callback listener.

⚠️ Authorized use only. This tool sends crafted requests that try to make a server reach unintended destinations. Run it only against systems you own or have explicit written permission to test. The CLI refuses to run without the --authorized flag.

Why

Server-side URL fetchers (webhooks, link previews, PDF/image renderers, import-by-URL features) are a common SSRF sink. Confirming the bug by hand is repetitive: you try loopback, the alternate IP encodings that slip past naive is it 127.0.0.1? checks, cloud metadata endpoints, non-HTTP schemes, and — when the response is blank — a blind callback. ssrf-probe packages that checklist so you can point it at an authorized target and get a short list to verify, instead of doing it one curl at a time.

What it covers

Category	Examples
Loopback / encodings	127.0.0.1, localhost, 2130706433, 0x7f.0x0.0x0.0x1, 0177.0.0.1, [::1], [::ffff:127.0.0.1]
Cloud metadata	AWS/OpenStack 169.254.169.254, GCP metadata.google.internal, Azure IMDS, Alibaba 100.100.100.200
Alternate schemes	file://, dict://, gopher://
Blind (out-of-band)	http://<token>.<your-oob-domain>/ confirmed by the built-in listener

Each cloud-metadata payload carries response markers (stable strings those endpoints return), so a reflected response is flagged likely automatically. A blind probe is only reported confirmed when your listener actually receives the callback.

Install

git clone https://github.com/yunshen-sec/ssrf-probe
cd ssrf-probe
pip install -e .

Pure standard library — no third-party runtime dependencies.

Usage

Reflected / direct SSRF against a lab target you control:

ssrf-probe --template 'http://target.local/fetch?url=FUZZ' --authorized

FUZZ marks where the target fetches a URL. Payloads are URL-encoded into that spot.

Blind SSRF with the out-of-band listener (run on a host the target can reach):

ssrf-probe \
  --template 'http://target.local/webhook?callback=FUZZ' \
  --authorized \
  --oob-domain oast.example.com \
  --listen 0.0.0.0:8080

Point the DNS for *.oast.example.com at the machine running the listener; each blind probe embeds a unique token so a hit is attributed to the exact payload.

Output

   [inconclusive] loopback http://127.0.0.1/ HTTP 200
>> [likely] cloud-metadata http://169.254.169.254/latest/meta-data/ HTTP 200 markers=ami-id,instance-id
>> [confirmed] oob http://<token>.oast.example.com/ OOB-HIT

>> lines (likely / confirmed) are the ones to verify by hand.

Library use

from ssrf_probe import Engine, all_payloads

engine = Engine()
for r in engine.run("http://target.local/fetch?url=FUZZ", all_payloads()):
    if r.confidence in {"likely", "confirmed"}:
        print(r.summary())

The Engine takes an injectable fetcher, so you can plug in your own HTTP client (auth headers, proxies) or run it fully offline in tests.

Development

pip install pytest
pytest -q

Responsible use

This is defensive/assessment tooling. Findings are meant to be reported to the system owner so the SSRF sink can be fixed (allow-list destinations, block link-local and metadata ranges, disable unused URL schemes). Do not run it against infrastructure you are not authorized to test.

License

MIT — see LICENSE.