@m8t-jacob/mcp-guard is a static, offline analysis tool: it reads .ts/.js source files and an optional JSON tools manifest from disk and pattern-matches against their text. It never executes the code it scans, never spawns the MCP server being scanned, and never makes outbound network calls of its own. The GitHub Action (action.yml) runs the same CLI inside your workflow's own runner.

The attack surface is therefore: malicious input designed to make the scanner itself misbehave (e.g. a pathological file crafted to cause catastrophic regex backtracking — a ReDoS), and the CLI's own argument parsing/file-reading logic.

mcp-guard being a security scanner does not mean its findings are exhaustive or authoritative — see the README's "Limitations" section for an honest account of what a regex-based analysis can and cannot catch. A missed vulnerability in a scanned server (a false negative) is a quality/ coverage issue, not a security vulnerability in mcp-guard itself — please report those as a regular bug (see .github/ISSUE_TEMPLATE/bug_report.md), not a security advisory.

If you discover an actual security vulnerability in mcp-guard itself (e.g. a ReDoS in one of its rule patterns triggerable by a crafted input file, or an issue in the CLI's own file/argument handling), please do not open a public issue. Instead, report it privately via GitHub Security Advisories for this repository.

Please include:

• A description of the vulnerability and its potential impact
• Steps to reproduce (a minimal input file/manifest that triggers it)
• The package version affected

We aim to acknowledge reports within 5 business days and to release a fix as soon as reasonably possible.