Your MCP Config Is the Actual Security Boundary. Nobody Audits It

Description and use

mcp-audit scans MCP server configurations for unsafe stdio commands, unauthenticated endpoints, open bind addresses, and known CVE versions, and can automatically apply fixes.

Problem

MCP server configurations are hand‑edited JSON files that define the security boundary, yet existing scanners only lint design or flag risky patterns without checking for known CVEs or providing fixes.

Outcome

mcp-audit identifies and remediates critical security issues in MCP configs, such as unsafe stdio commands, unauthenticated endpoints, and vulnerable package versions, improving the security posture of locally run MCP servers.

Workflow

• 1. Install the tool: pip install mcp-runtime-audit mcp-audit scan
• 2. Run a scan: mcp-audit scan (optionally with --probe to test endpoints)
• 3. Review findings in text or JSON output
• 4. Apply fixes: mcp-audit fix --apply (dry‑run with --dry-run first)
• 5. Optionally link agent‑guard: mcp-audit scan --link-agent-guard

Evidence

So I built mcp-audit — a local runtime security scanner for MCP servers with a bundled CVE table and one‑click fixes. stdlib only, no dependencies.

Source

Every finding carries a rule id, CVE id, severity, the evidence, and the fix — in text and JSON output, with CI‑ready exit codes (0 clean, 1 findings, 2 errors).

Source

The tool rewrites 0.0.0.0 → 127.0.0.1 in URLs, flags, and env values, adds an Authorization: Bearer ${MCP_TOKEN} placeholder to unauthenticated endpoints, backs up the config to <file>.bak‑<timestamp> before writing anything.

Source