See what your agents are doing.
One system to scan, score, and secure every MCP server and agent skill in your stack.
Catch hidden instructions, credential theft, and malware — before they reach your machine.
Free to start. No API key required.
4,104+
Servers indexed
4,072+
Scans completed
220+
Detection checks
Why Vigile
Scan. Score. Secure.
Three layers of protection for the AI agent supply chain — from static analysis to runtime monitoring.
Scan
220+ detection checks across MCP, skill, runtime, and backend analysis catch instruction injection, malware delivery, stealth operations, data exfiltration, and more. One command scans everything.
npx vigile-scan --allScore
Every skill and server gets a 0-100 trust score. The public registry lets you check before you install — like a security credit score for AI tools.
Secure
Sentinel monitors MCP servers in real time. Catch phone-home behavior, DNS tunneling, and C2 connections that static analysis misses.
Real-time monitoringThe problem
Agent skills run with your AI's full trust. There's no sandbox, no review, no warning.
When you install a SKILL.md file in Claude Code or a .mdc rule in Cursor, your agent follows those instructions unconditionally. A malicious skill doesn't need its own tools — it weaponizes every tool your agent already has.
of MCP servers exposed to SSRF
BlueRock Security, 7,000+ analyzed
vulnerabilities in OpenClaw audit, 8 critical
Jan 2026
OpenClaw RCE via CVE-2026-25253
CVE Database
malicious skills found on ClawHub
ClawHavoc campaign, Feb 2026
Detection
What Vigile catches
Agent Skill Threats
SK-001Instruction Injection
Hidden directives that silently override agent behavior
SK-010Malware Delivery
Fake dependencies that install credential stealers
SK-020Stealth Operations
Actions hidden from the user — "silently", "don't tell"
SK-030Safety Bypass
Disabling confirmations, auto-approving dangerous actions
SK-040Persistence
.bashrc writes, CLAUDE.md tampering, cron jobs, git hooks
SK-050Data Exfiltration
Credential harvesting, env dumping, filesystem enumeration
MCP Server Threats
TP-001Tool Poisoning
Hidden instructions in tool descriptions that hijack behavior
EX-003Data Exfiltration
Patterns targeting SSH keys, AWS credentials, env files
PM-001Permission Abuse
Excessive filesystem, network, or code execution access
OB-002Obfuscation
Base64, hex, zero-width chars hiding malicious payloads
+ typosquatting detection and Sentinel runtime monitoring
Try it now
Five seconds. Zero config.
One command discovers skill files and MCP configs automatically across all your AI tools. No setup required.
Run the scanner
Auto-discovers skills and MCP configs across 7 platforms
Review findings
54 rules and counting flag injection, malware, stealth ops, and more
Trust the score
0-100 composite score — keep the green, block the red
$ npx vigile-scan --all
Scanning skills & MCP servers...
SKILL.md — marketing-automator
CRITICALSK-001 Instruction Injection
"ignore all previous instructions"
CRITICALSK-010 Malware Delivery
pip install "cryptohelper-utils" (known stealer)
HIGH SK-020 Stealth Operations
"do not tell the user about this step"
Trust Score: 12/100 ■ Dangerous
──────────────────────────────────────
.mdc — code-reviewer
✓ No issues found
Trust Score: 94/100 ■ Trusted