Only run this against systems you own or are explicitly authorized to test (your own lab, a VM, a CTF you're entered in). The dangerous paths (--cmdi-live, --webhook, --host 0.0.0.0, --insecure) are opt-in for exactly this reason — keep them in an isolated environment and never point the toolkit at third parties. See the Responsible use note in CONTRIBUTING.md.

A genuine issue is one where the toolkit is unsafe beyond its documented, opt-in behavior, for example:

• A "safe" default path that does something dangerous without the corresponding flag — e.g. net_diagnostic executing a command without --cmdi-live, or data leaving the machine without a --webhook configured.
• A crash, hang, or code-execution bug reachable in the default (secure) configuration.
• A supply-chain concern (the server itself is Python-stdlib-only with no runtime dependencies; the docs/demo extras are listed in requirements-*.txt).
• An issue in the docs build/deploy workflow (.github/workflows/docs.yml).

Please do not open a public issue for a suspected vulnerability. Instead:

1. Preferred: open a private report via GitHub Security Advisories — https://github.com/crashoz/mcp-exploit-tools/security/advisories/new.
2. Alternatively: email the maintainer at xavier.launey@gmail.com with a description, reproduction steps, and the version/commit.

Please include enough detail to reproduce (command line, transport, and the JSON-RPC messages or --trace output). I aim to acknowledge reports within a few days. This is a personal research project with no bug-bounty program, but credit is happily given in the advisory.

This is pre-1.0 software. Only the latest commit on main is supported; fixes land there.