mcp-fortify

Security scanner for MCP (Model Context Protocol) configurations. Scans your local MCP setup for hardcoded secrets, insecure permissions, command injection risks, and missing security hooks — fully offline, no API keys needed.

Why

43% of MCP implementations have command injection vulnerabilities (Equixly research). While tools like mcp-scan analyze MCP server source code, no good static scanner exists for checking whether YOUR local MCP installation is configured securely.

mcp-fortify answers: "Is MY MCP setup secure?"

Quick Start

npx mcp-fortify

Or install globally:

npm install -g mcp-fortify
mcp-fortify

What It Scans

Auto-discovers and scans these files on your machine:

| File | Location | |------|----------| | Claude Code settings | ~/.claude/settings.json | | Claude Desktop config | ~/Library/Application Support/Claude/claude_desktop_config.json | | MCP server configs | ~/.claude/mcp-servers/*/ (run.sh, .env, server.py, etc.) | | Project config | .mcp.json in current directory |

Security Rules (8 rules)

| Rule | Severity | What It Detects | |------|----------|-----------------| | hardcoded-secrets | CRITICAL | API keys, tokens, passwords in any scanned file | | plaintext-env | CRITICAL | Secrets in JSON config env blocks | | command-injection | CRITICAL | Shell injection vectors in launch scripts and configs | | file-permissions | HIGH | Config/credential files not restricted to owner-only | | transport-security | HIGH | HTTP (not HTTPS) endpoints, 0.0.0.0 bindings | | missing-hooks | MEDIUM | No PreToolUse hooks for secret blocking in Claude Code | | tool-permissions | MEDIUM | Wildcard tool permissions bypassing safety checks | | missing-gitignore | MEDIUM | .env files not covered by .gitignore |

Commands

mcp-fortify scan (default)

Scan MCP configurations for security issues.

mcp-fortify                              # Scan with defaults
mcp-fortify --format json                # JSON output for CI
mcp-fortify --format sarif               # SARIF for GitHub Security tab
mcp-fortify --format html -o report.html # HTML report
mcp-fortify --severity critical --ci     # Fail CI on critical findings
mcp-fortify --rules hardcoded-secrets    # Run specific rules only
mcp-fortify --custom-rules ./my-rules.js # Load custom rules
mcp-fortify --verbose                    # Show all scanned files
mcp-fortify scan /path/to/project        # Scan custom path

mcp-fortify init

Generate security hooks for Claude Code. Creates a PreToolUse hook that blocks file writes containing secrets.

mcp-fortify init

mcp-fortify fix

Auto-remediate fixable issues (file permissions).

mcp-fortify fix             # Fix issues
mcp-fortify fix --dry-run   # Preview what would be fixed

Output Formats

| Format | Use Case | |--------|----------| | console | Terminal output with colors (default) | | json | CI pipelines, programmatic consumption | | sarif | GitHub Security tab, CodeQL integration | | html | Shareable reports, audits |

CLI Options

mcp-fortify [command] [options]

Commands:
  scan [path]     Scan MCP configurations (default)
  init            Generate security hooks for Claude Code
  fix             Auto-fix remediable security issues

Scan Options:
  -f, --format <type>       Output format: console | json | sarif | html (default: console)
  -o, --output <file>       Write output to file instead of stdout
  -s, --severity <level>    Minimum severity to show (critical|high|medium|low|info)
  --rules <ids>             Comma-separated rule IDs to run
  --custom-rules <path>     Path to custom rules module (.js)
  --no-color                Disable colors
  --verbose                 Show all scanned files
  --ci                      Exit code 1 if high+ severity findings

Fix Options:
  --dry-run                 Show what would be fixed without making changes

Custom Rules

Create a .js file that exports an array of rule objects:

// my-rules.js
export default [
  {
    id: 'my-custom-rule',
    name: 'My Custom Rule',
    severity: 'high',
    description: 'Checks for something specific',
    run(targets) {
      const findings = [];
      for (const target of targets) {
        // Your detection logic here
      }
      return findings;
    },
  },
];

Then run: mcp-fortify --custom-rules ./my-rules.js

Safe Patterns (Not Flagged)

mcp-fortify recognizes secure patterns and won't flag them:

• $(secrets get KEY_NAME) — macOS Keychain references
• $(gh auth token) — GitHub CLI auth
• ${ENV_VAR} — Shell variable expansion
• process.env.KEY — Node.js env references
• Placeholder values (YOUR_KEY_HERE, changeme, etc.)

Programmatic API

import { scan, format } from 'mcp-fortify';

const result = await scan({ format: 'json' });
console.log(result.findings);

// Generate HTML report
const html = format(result, 'html');

License

MIT