vigile-scan

Security scanner for MCP servers and AI agent skills. Detect tool poisoning, credential theft, data exfiltration, and supply chain attacks — before they reach your machine.

Quick Start

npx vigile-scan

No install, no config. Vigile discovers your MCP server configurations and agent skill files, scans them against 59 detection rules, and returns a trust score for each one.

What It Detects

MCP Server Threats (22 patterns + 5 inline checks)

Agent Skill Threats (32 patterns)

Platforms

Vigile auto-discovers configurations from:

• Claude Desktop — claude_desktop_config.json
• Claude Code — CLAUDE.md, .claude/ skill files
• Cursor — .cursor/rules/*.mdc, .cursorrules
• GitHub Copilot — .github/copilot/**/*.md, copilot-instructions.md
• Windsurf — windsurf.json, .windsurfrules
• VS Code — .vscode/mcp.json
• OpenClaw — ~/.openclaw/openclaw.json, openclaw.config.json

Usage

vigile-scan [options]

Scan Options

Sentinel Runtime Monitoring (Pro)

Authentication

# Authenticate with your API key (get one at https://vigile.dev/account)
vigile-scan auth login <vgl_your_api_key>

# Check auth status
vigile-scan auth status

# Log out
vigile-scan auth logout

You can also set VIGILE_TOKEN as an environment variable for CI/CD.

Examples

Scan everything

npx vigile-scan --all

JSON output for CI/CD

npx vigile-scan --json --all > vigile-report.json

Scan a specific client

npx vigile-scan --client cursor

Verbose output with score breakdown

npx vigile-scan --all --verbose

GitHub Actions

- name: Vigile Security Scan
  run: npx vigile-scan --all --json -o vigile-report.json
  env:
    VIGILE_TOKEN: ${{ secrets.VIGILE_TOKEN }}

- name: Fail on critical findings
  run: |
    critical=$(jq '.bySeverity.critical' vigile-report.json)
    if [ "$critical" -gt 0 ]; then exit 1; fi

A dedicated GitHub Action (vigile-github-action) is also available for deeper CI/CD integration.

Trust Scores

Every scanned item gets a trust score from 0–100:

The score is a weighted composite of five factors: code analysis (30%), dependency health (20%), permission safety (20%), behavioral stability (15%), and transparency (15%).

Exit Codes

This makes vigile-scan work natively in CI/CD pipelines — a non-zero exit fails the build.

Sentinel: Runtime Monitoring

Static scanning catches what's in the code. Sentinel catches what the code actually does on the wire.

When you run --sentinel, Vigile intercepts outbound network traffic from your MCP servers and flags:

• C2 beaconing — periodic callbacks to unknown servers
• Credential theft — API keys, tokens, or secrets sent over the network
• DNS tunneling — data exfiltration hidden in DNS queries
• Unexpected destinations — connections to IPs/domains outside the expected set

Sentinel is available on Pro ($30/mo) and Pro+ ($100/mo) plans. Free users can run static scans with no limits.

Pricing

Annual plans available at vigile.dev/pricing.

Links

• Web Scanner & Registry — vigile.dev
• GitHub — github.com/Vigile-ai/vigile-scan
• Report Issues — github.com/Vigile-ai/vigile-scan/issues

License

Apache-2.0