Skip to main content
Glama

Security Knowledge MCP Server

A knowledge-focused MCP (Model Context Protocol) server that guides AI agents through security assessments. Instead of executing tools directly, it provides methodology, documentation, and step-by-step workflows - letting your AI agent run the CLI commands with full context.

Installation

npx github:yunusj/VibeDefender-MCP

This automatically clones, installs dependencies, builds, and runs the MCP server.

Global Installation

npm install -g github:yunusj/VibeDefender-MCP vibedefender-mcp

Philosophy

The MCP GUIDES, not executes.

Your AI agent (Claude Code, Cursor, etc.) already has the ability to run CLI commands. This MCP provides:

  • Step-by-step methodology for security assessments

  • Installation guides for required tools

  • Plain English explanations for non-technical users

  • Documentation proxy for always-current tool docs

Configuration

Claude Code

Add to your Claude Code MCP settings:

{ "mcpServers": { "vibedefender": { "command": "npx", "args": ["github:yunusj/VibeDefender-MCP"] } } }

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{ "mcpServers": { "vibedefender": { "command": "npx", "args": ["github:yunusj/VibeDefender-MCP"] } } }

Cursor

Add to Cursor's MCP settings (Settings → Features → MCP):

{ "mcpServers": { "vibedefender": { "command": "npx", "args": ["github:yunusj/VibeDefender-MCP"] } } }

Google Antigravity

Same configuration as Claude Desktop.

Testing with MCP Inspector

Test the server before configuring in your editor:

npm install -g @modelcontextprotocol/inspector npx @modelcontextprotocol/inspector npx github:yunusj/VibeDefender-MCP

Expected: Web UI shows "Connected", lists all security:// resources and 5 workflow prompts.

Quick Start

Just say what you want in plain English:

  • "Help me set up security scanning" - Guides you through installing tools

  • "Scan my code for security issues" - Runs comprehensive scans

  • "Check my code before I push" - Quick pre-commit check

  • "Test my app on localhost" - Starts dev server and runs live tests

  • "Check this URL for vulnerabilities" - Tests a specific URL

Available Prompts

Prompt

What It Does

setup

Checks what tools are installed, guides installation

scan

Comprehensive scan: static analysis + runtime testing + artifacts

pre-push

Quick critical-only check (fast, no runtime analysis)

scan-live

Starts your dev server and runs live security tests

scan-url

Tests a URL you provide (with authorization check)

Required Tools

The MCP guides you through installing these (just say "help me set up"):

Tool

Purpose

Install (macOS)

Trivy

CVE/dependency scanning

brew install aquasecurity/trivy/trivy

Semgrep

Static code analysis

brew install semgrep

Nuclei

Runtime testing (mandatory)

brew install nuclei

Metasploit Integration

Metasploit Framework is integrated for both discovery (reconnaissance) and exploitation phases.

Setup

  1. Install external Metasploit MCP server:

# Clone the Metasploit MCP repository git clone https://github.com/your-org/MetasploitMCP ~/MetasploitMCP
  1. Set environment variable (add to ~/.bashrc or ~/.zshrc):

export METASPLOIT_MCP_PATH="$HOME/MetasploitMCP/start_mcp.sh"
  1. Verify installation:

npm run mcp:metasploit # Should output: "Metasploit MCP Proxy running on stdio"

Usage

Discovery Phase (Automatic):

  • Runs safe auxiliary modules for service detection

  • Port scanning and version detection

  • Correlates findings with CVE database

  • No exploitation attempts

Exploitation Phase (Requires Approval):

  • Executes exploits against validated vulnerabilities

  • Requires explicit human approval

  • Full session management and post-exploitation

  • Evidence gathering only (read-only)

Safety

⚠️ CRITICAL: Authorization Required

  • Only use on systems you own OR have written permission to test

  • Unauthorized access to computer systems is illegal

  • Discovery phase: Basic authorization sufficient

  • Exploitation phase: Explicit approval required

Documentation

  • See security://methodology/exploitation for complete exploitation workflow

  • See security://docs/metasploit for detailed tool usage

  • Includes safety guidelines, workflow examples, and troubleshooting

Available Resources

Methodology Guides

Resource

Description

security://methodology/discovery

How to find attack surface

security://methodology/scanning

What to scan, in what order

security://methodology/validation

How to validate on live targets

security://methodology/reporting

How to structure reports

Reference Documentation

Resource

Description

security://reference/severity

What CRITICAL/HIGH/MEDIUM/LOW mean

security://reference/cwe-top-25

Common weakness types

security://reference/owasp-top-10

OWASP vulnerability categories

security://reference/browser-options

Browser automation choices

Tool Documentation

Resource

Description

security://docs/overview

Overview of all tools

security://docs/trivy

Trivy quick start

security://docs/semgrep

Semgrep quick start

security://docs/nuclei

Nuclei quick start

security://docs/playwright

Playwright MCP info

security://docs/metasploit

Metasploit basics

Example Workflows

First Time Setup

You: Help me set up security scanning AI: I'll check what tools you have installed... [Runs version checks] You're missing Trivy and Semgrep. Let me help you install them. Trivy checks your dependencies for known vulnerabilities. Run this to install: brew install aquasecurity/trivy/trivy ...

Before Pushing Code

You: Check my code before I push AI: Running quick pre-push security check... [Runs trivy + semgrep for CRITICAL issues only] ✅ All clear! No critical security issues found. Safe to push. For a comprehensive scan including runtime testing, use the 'scan' workflow.

Full Security Scan

You: Scan my code for security issues AI: I see this is a Node.js project using Express. First, checking for known vulnerabilities in dependencies... [Runs trivy] Found 3 critical issues in your packages: 1. lodash < 4.17.21 - allows prototype pollution Fix: npm update lodash ...

Project Structure

src/mcp-server/ ├── index.ts # Main server (resources + prompts only) ├── resources/ │ ├── static.ts # Bundled methodology & references │ └── dynamic.ts # Tool documentation with search guidance └── prompts/ └── workflows.ts # 5 user-friendly workflow prompts

Design Decisions

  1. No execution tools - AI agents already have CLI access. We provide knowledge.

  2. Plain English - Everything explained for non-technical users.

  3. Step-by-step - Prompts tell the AI exactly what to do at each step.

  4. Documentation proxy - Search patterns for always-current tool docs.

  5. Minimal files - 4 files total, easy to understand and maintain.

  6. GitHub-based distribution - No npm publish, direct from source via npx.

Browser Automation Options

For live testing that needs a browser:

Option

When to Use

Playwright MCP

Claude Code, Cursor, most IDEs

Browser Agent

Google Antigravity IDE (built-in)

Puppeteer

If already in project

Troubleshooting

Build Errors

If you see TypeScript compilation errors when installing from GitHub:

npm cache clean --force npx github:yunusj/VibeDefender-MCP

MCP Server Not Connecting

  1. Test with MCP Inspector first (see "Testing" section above)

  2. Check Node.js version: node --version (requires >= 22.0.0)

  3. Verify the server runs standalone:

    npx github:yunusj/VibeDefender-MCP # Should output: "Security Knowledge MCP server running on stdio"
  4. Check editor configuration file syntax (valid JSON)

  5. Restart your AI editor after configuration changes

Permission Errors

If you get EACCES errors:

# On Unix-like systems, the shebang should make it executable # If not, manually set permissions on global install: chmod +x $(which vibedefender-mcp)

Update to Latest Version

npm cache clean --force npx github:yunusj/VibeDefender-MCP # Or for global install npm uninstall -g vibedefender-mcp npm install -g github:yunusj/VibeDefender-MCP

Security Notice

  • Only scan systems you are authorized to test

  • Live validation (scan-url, scan-live) requires explicit authorization

  • The AI will ask for confirmation before testing URLs

  • Always get written permission before security testing

  • Follow responsible disclosure practices

License

MIT

-
security - not tested
A
license - permissive license
-
quality - not tested

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/yunusj/VibeDefender-MCP'

If you have feedback or need assistance with the MCP directory API, please join our Discord server