● For developers shipping MCP servers & AI agents

Ship an MCP server the internet can't turn into a file-read or RCE machine.

Every tool your LLM calls is driven by untrusted content. One missed ../ or shell=True and your agent leaks /etc/passwd or runs attacker commands. This kit is the exact checklist, copy-paste guards, and attacker payloads that stop it — pulled from real audits of production MCP servers.

Get the Kit — $19
One-time. Instant access. Written by a working security researcher.

What's inside

Path confinement
The #1 MCP CVE class. A symlink- and absolute-path-proof resolve_within() and the call-site regression that re-opens it.
No-shell command exec
Argv-array pattern + how to block argument injection on the args users do influence.
Deserialization (RCE)
pickle / torch.load / yaml / numpy — huntr's most-paid class — and the safe replacements.
SSRF in URL fetchers
Block cloud-metadata & localhost with a private-IP resolver you can paste in.
Attacker test payloads
The exact strings to fire at your own server so you find the hole before they do.
CI pre-deploy checklist
Grep one-liners + a box-tick list to run every release.

Free sample — the pre-deploy audit checklist

[ ] grep -rn "shell=True\|os.system\|subprocess.call(" src/
[ ] grep -rn "pickle.load\|yaml.load(\|torch.load(\|allow_pickle=True" src/
[ ] grep -rn "open(\|Path(\|send_file\|extractall" src/   # every path sink
[ ] every path sink flows through resolve_within() and refuses None
[ ] every subprocess is an argv list with a timeout; user args whitelisted
[ ] every URL fetch is scheme+IP validated, no redirects, size-capped
[ ] no secrets in code/output; inputs length+type bounded
[ ] ran the attacker payloads against a LOCAL instance — all refused
The full kit gives you the code that makes every one of these boxes pass — copy-paste, framework-agnostic Python.

Why trust it

Written by a security researcher who audits MCP servers and AI/ML packages for a living. Every guard here maps to a real CVE class (e.g. path-traversal file-read in MCP file tools, CVE-2026-54785). No theory, no AI slop — patterns that shipped.

Get the Kit — $19

Secure Stripe checkout. Not happy? Reply to your receipt for a refund.
Get the Kit — $19 →