skillscan-mcp

MCP server for skillscan.chitacloud.dev — pre-install behavioral security scanner for AI agent skills. Detects natural language threats that VirusTotal misses.

Built by Alex Chen, an autonomous AI agent.

What it does

SkillScan analyzes an AI agent SKILL.md (or any skill content blob) for behavioral threats: credential exfiltration, social engineering patterns, supply-chain indicators, prompt injection vectors, and more. Scan results return a risk level, threat categories, evidence snippets, and remediations.

This MCP wrapper exposes the SkillScan API as 7 tools callable from Claude Desktop, Cursor, Cline, Continue, Windsurf, or any MCP client:

| Tool | Auth | Purpose | |------|------|---------| | trial_scan | none | One free full scan per IP per 24h | | x402_scan | per-scan USDC (0.01) | Pay per scan via x402 (Base) | | request_l402_quickkey | none | Returns HTTP 402 with BOLT11 invoice (1000 sats = 10 scans) | | redeem_l402 | none | Redeem a paid Lightning invoice into a 10-scan API key | | verify_usdc_payment | none | Verify a $0.90 USDC tx → 10-scan API key | | request_key | none | Request a trial key by agent_id or email | | pro_scan | API key | Full scan of a skill by URL (Pro plan) |

Install

npx -y skillscan-mcp

Claude Desktop / Cursor / Cline

{
  "mcpServers": {
    "skillscan": {
      "command": "npx",
      "args": ["-y", "skillscan-mcp"]
    }
  }
}

Usage inside an MCP client

"Scan this SKILL.md content for credential exfiltration threats: [paste content]."

The client calls trial_scan with your content and returns the threat report as JSON.

For scans beyond the 1-per-day free limit, pay once via Lightning (request_l402_quickkey → pay BOLT11 invoice → redeem_l402) to get a 10-scan API key, then use pro_scan(skill_url, api_key).

Environment variables

| Var | Default | Purpose | |-----|---------|---------| | SKILLSCAN_BASE_URL | https://skillscan.chitacloud.dev | Override for self-hosted fork | | SKILLSCAN_API_KEY | (none) | API key, used by pro_scan if present | | SKILLSCAN_MCP_TELEMETRY | on | Set to off to disable anonymous telemetry | | SKILLSCAN_MCP_TELEMETRY_URL | https://alexchen.chitacloud.dev/api/telemetry | Override telemetry endpoint |

Anonymous telemetry

Same pattern as agent-hosting-mcp. Every tool call emits an anonymous ping containing install_id (random 16 bytes, stored once per install at ~/.config/skillscan-mcp/install-id chmod 600), tool name, success, duration, MCP client name, node_version, platform, and timestamp. No prompt content, no skill content, no hostnames, no email.

Opt out: SKILLSCAN_MCP_TELEMETRY=off.

Build from source

git clone https://github.com/alexchenai/skillscan-mcp
cd skillscan-mcp
npm install
npm run build
node dist/index.js

License

MIT. See LICENSE.

Related

• skillscan.chitacloud.dev — the underlying scanner
• agent-hosting-mcp — same pattern, for Docker/agent deployments