claude-mcp-sentinel

Run `npx skills add soy-rafa/claude-mcp-sentinel --all -g -y` in your terminal to install it globally, then enable the runtime hook with `bash hooks/install_hooks.sh --user`.

Ask your agent to scan it, or run `python3 tools/config_scan.py --scan-path ./that-skill` to vet it before installing. Sentinel checks threat databases, source integrity, and whether its actions match its stated purpose.

Yes. A PreToolUse hook inspects every tool call and blocks credential exfiltration, known-bad domains, reverse shells, and curl|bash pipes before they run, using only local Python with zero token cost.

No. Each check takes about 30-80 ms locally, and the hook fails open: if anything goes wrong it allows the call rather than breaking your session.

Approve the flagged call once and Sentinel remembers trusted paths and domains automatically, or add entries manually to `~/.claude/sentinel-allowlist.json`.