Cyber Lens AI MCP Server

Security scanning for AI assistants and agentic coding workflows.

A Model Context Protocol (MCP) server that gives AI assistants the ability to scan websites, public repositories, and Open CLAW skills for security vulnerabilities. It's purpose-built for the agentic coding era, with native support for scanning skill packages before you install them.

What It Does

When connected to an AI assistant (Claude Desktop, VS Code, etc.), this MCP server adds security scanning tools that the assistant can use during your conversation:

• Scan a CLAW skill before installing it -- the server downloads the skill package, extracts it, and analyses the actual source code for dangerous patterns, hardcoded secrets, and permission issues
• Scan a website in a local quick mode without an account, or use the full cloud scan when connected
• Scan a public repository for exposed secrets, dependency vulnerabilities, suspicious code patterns, and trust posture issues
• Check your remaining cloud scan quota before kicking off a batch of scans
• Get remediation guidance with step-by-step fix instructions and code examples

Example Conversation

You:    "I want to install this skill: https://clawhub.ai/skills/ontology"

Claude: I'll scan it for security issues first.

        [PASS] CLAW Skill Security Scan: ontology v1.0.4

        Security Score: 95/100
        Files Analyzed: 5

        Skill appears safe to install. Minor issues noted below.

        1. [MEDIUM] File write operation
           File: scripts/ontology.py:399
           Fix: Ensure file writes are limited to expected directories.

        The skill looks safe. It writes to the filesystem for its knowledge
        graph storage, which is expected behaviour for this type of skill.

Getting Started

1. Install

After the first npm release, the recommended install path is:

npx -y @shadoprizm/cyberlens-mcp-server

Before that, or for local development and manual builds:

git clone https://github.com/shadoprizm/cyberlens-mcp-server.git
cd cyberlens-mcp-server
npm install
npm run build

2. Add to Your AI Assistant

Claude Desktop -- add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "cyberlens": {
      "command": "npx",
      "args": ["-y", "@shadoprizm/cyberlens-mcp-server"]
    }
  }
}

If you are running from a source checkout instead of npm:

{
  "mcpServers": {
    "cyberlens": {
      "command": "node",
      "args": ["/absolute/path/to/cyberlens-mcp-server/dist/index.js"]
    }
  }
}

Claude Code -- add to your MCP settings or project configuration.

VS Code (Copilot/Continue) -- configure per your extension's MCP server settings.

3. Connect Your Account

Website tools work immediately without an account in local quick mode. That local mode covers roughly 15 core checks and returns results right away.

Connecting an account upgrades website scans to the full CyberLens cloud path with 70+ checks, scan history, and AI analysis. Repository and account-only tools still connect through the browser flow when needed.

When a repository or account-only tool needs an account, the MCP server:

1. opens the CyberLens browser flow automatically
2. sends the user to cyberlensai.com to sign up or log in
3. receives the secure callback locally
4. saves the API key to ~/.cyberlens/mcp/config.json
5. continues the original tool call automatically

You can also trigger the same flow explicitly:

You: "Connect my CyberLens account"

This opens your browser to cyberlensai.com where you can sign up (free) or log in. Your API key is saved locally at ~/.cyberlens/mcp/config.json and used for all future scans.

Free accounts include 5 scans/month. No credit card required.

You can also set the CYBERLENS_API_KEY environment variable in the MCP config instead of using the browser flow.

If a cloud website scan hits its monthly quota, the MCP server opens the CyberLens pricing page automatically and falls back to the local quick scan instead of hard-failing. Repository scans still require cloud quota.

Available Tools

Account

CLAW Skill Scanning

Website & Repository Scanning

Intelligence & Guidance

How Skill Scanning Works

When you provide a CLAW skill URL, the MCP server:

1. Resolves the URL -- accepts Claw Hub pages (https://clawhub.ai/skills/skill-name), direct download links (https://*.convex.site/api/v1/download?slug=name), or any URL pointing to a skill zip
2. Downloads the skill package to a temporary directory
3. Extracts and analyses every file in the package:
   • Checks _meta.json and SKILL.md for completeness
   • Scans all source code (Python, JavaScript, TypeScript, shell scripts, config files) for dangerous patterns
   • Detects hardcoded secrets, eval() usage, shell command execution, insecure HTTP requests, file deletion operations, pickle deserialisation, and more
   • Identifies unpinned dependencies
   • Catalogues all external URLs the skill communicates with
4. Returns a security score (0-100) with detailed findings, file locations, and fix recommendations
5. Cleans up all temporary files

This entire process runs locally -- no API key is required and your code is never sent to an external server.

Website Scan Modes

scan_website and get_security_score now have two honest modes:

• Local Quick Scan -- works without an account, returns immediately, and covers roughly 15 core website checks such as HTTPS, security headers, server disclosure, insecure forms, and inline-script indicators
• Full Cloud Scan -- requires a connected CyberLens account, runs 70+ checks, keeps cloud scan history, and includes richer analysis

If a user asks for a full or database website scan without an account, the MCP server still returns the local quick scan and says that the requested cloud-only mode was not available.

If a connected user runs out of website cloud quota, CyberLens falls back to the local quick scan automatically and opens the pricing page with an upgrade link.

Accepted URL Formats

https://clawhub.ai/skills/ontology            --> Resolves automatically
https://clawhub.ai/author/skill-name          --> Resolves automatically
https://*.convex.site/api/v1/download?slug=x  --> Direct download

Architecture

src/
  index.ts          MCP server, tool handlers, output formatting
  auth.ts           Browser-based connect flow, config file management
  client.ts         REST API client for live scan and quota endpoints
  remediation-guides.ts  Local CWE and vulnerability remediation guidance
  schemas.ts        Zod input validation schemas
  skill-scanner.ts  Local CLAW skill analyser (download, extract, scan)
  skill-validation.ts    Local CLAW manifest validation
  transparency.ts        Local transparency report for scan coverage

Key design decisions:

• No Supabase SDK -- pure REST calls with fetch and X-API-Key header
• Truthful cloud surface -- the MCP server only exposes cloud-backed tools that are supported by the live public API (/scan, /scan/{id}, /quota)
• Stdio transport -- runs as a subprocess of the AI assistant, communicates via stdin/stdout
• Useful without API key -- skill scanning, website quick scans, manifest validation, remediation guidance, and transparency reporting work locally; connecting an account upgrades website scans to the full cloud path and unlocks repository scanning
• Browser-based auth -- same secure connect flow as the CyberLens OpenClaw skill (CSRF-protected, short-lived exchange codes, HTTPS-only)

Development

# Build
npm run build

# Watch mode
npm run dev