Code Audit MCP Server

A Model Context Protocol (MCP) server providing comprehensive code security analysis and dependency auditing. Performs Static Application Security Testing (SAST), dependency vulnerability scanning, secret/credential detection, and Software Bill of Materials (SBOM) generation.

Overview

The Code Audit MCP server integrates industry-standard security scanning tools into Claude, enabling AI-assisted code auditing. It combines multiple security disciplines:

• SAST Analysis: Semgrep pattern-based vulnerability detection
• Dependency Auditing: npm, pip, and cargo vulnerability checking
• Secret Detection: Credential and API key discovery via gitleaks or trufflehog
• SBOM Generation: Software composition tracking with CycloneDX/SPDX formats

All scanning operates within a secure, path-validated environment with timeout protection.

Tools

Available Security Scanning Tools

| Tool | Purpose | Input | Output | |------|---------|-------|--------| | semgrep_scan | Run Semgrep SAST rules against source code | Directory path, config profile | Security findings with severity and location | | dep_audit | Check dependencies for known vulnerabilities | Directory path, package manager | CVEs, versions, and severity levels | | secret_scan | Detect secrets and credentials in code | Directory path, max results | Matched credential patterns with file location | | sbom_generate | Generate Software Bill of Materials | Directory path, format | Component inventory with versions and licenses |

Tool Details

semgrep_scan

Runs Semgrep rules to identify security vulnerabilities and code quality issues using pattern-based static analysis.

Input Schema:

{
  "target": "string (required)",
  "config": "string (default: 'auto')",
  "max_findings": "integer (default: 100, range: 1-500)"
}

Parameters:

• target: Absolute path to project directory to scan
• config: Semgrep configuration to use. Options include:
  • auto (default) - Uses Semgrep's default rules
  • p/security-audit - Security-focused ruleset
  • p/owasp-top-ten - OWASP Top 10 vulnerabilities
  • p/cwe-top-25 - CWE Top 25 weaknesses
  • Custom registry path or local rulefile
• max_findings: Maximum number of findings to return (default: 100)

Example Request:

{
  "target": "/home/user/projects/webapp",
  "config": "p/security-audit",
  "max_findings": 50
}

Example Output:

{
  "target": "/home/user/projects/webapp",
  "config": "p/security-audit",
  "total_findings": 12,
  "findings": [
    {
      "rule_id": "python.django.security.injection.sql_injection.sql_string_formatting",
      "severity": "HIGH",
      "message": "Potential SQL injection via string formatting",
      "file": "/home/user/projects/webapp/models.py",
      "line_start": 45,
      "line_end": 45,
      "code_snippet": "User.objects.raw(f\"SELECT * FROM users WHERE id={user_id}\")"
    },
    {
      "rule_id": "python.django.security.injection.xss.template_tag_xss",
      "severity": "MEDIUM",
      "message": "Potential XSS vulnerability in template",
      "file": "/home/user/projects/webapp/templates/profile.html",
      "line_start": 12,
      "line_end": 12,
      "code_snippet": "{{ user_input | safe }}"
    },
    {
      "rule_id": "python.django.security.hardcoded_secrets.hardcoded_secret_in_settings",
      "severity": "CRITICAL",
      "message": "Hardcoded secret key detected",
      "file": "/home/user/projects/webapp/settings.py",
      "line_start": 78,
      "line_end": 78,
      "code_snippet": "SECRET_KEY = 'django-insecure-a1b2c3d4e5f6g7h8i9j0k1l2m3n4o5p6'"
    }
  ],
  "errors": []
}

dep_audit

Audits project dependencies for known security vulnerabilities. Automatically detects package manager (npm, pip, cargo) or accepts explicit specification.

Input Schema:

{
  "target": "string (required)",
  "package_manager": "string (default: 'auto')"
}

Parameters:

• target: Absolute path to project directory containing dependency files
• package_manager: Package manager to audit. Options:
  • auto (default) - Auto-detect from lock files (package-lock.json, requirements.txt, Cargo.lock)
  • npm - Node.js packages (npm audit)
  • pip - Python packages (pip-audit)
  • cargo - Rust packages (cargo audit)

Detection Logic:

• npm: Looks for package-lock.json or package.json
• pip: Looks for requirements.txt, setup.py, or pyproject.toml
• cargo: Looks for Cargo.lock or Cargo.toml

Example Request:

{
  "target": "/home/user/projects/api-server",
  "package_manager": "auto"
}

Example Output (npm vulnerabilities):

{
  "target": "/home/user/projects/api-server",
  "package_manager": "npm",
  "total_vulnerabilities": 3,
  "vulnerabilities": [
    {
      "package_name": "express",
      "severity": "HIGH",
      "title": "Express open redirect vulnerability",
      "installed_version": "4.17.1",
      "patched_version": "4.17.3",
      "cve": "CVE-2022-24999",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2022-24999"
    },
    {
      "package_name": "lodash",
      "severity": "MEDIUM",
      "title": "Lodash isEqual prototype pollution",
      "installed_version": "4.17.20",
      "patched_version": "4.17.21",
      "cve": "CVE-2021-23337",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2021-23337"
    },
    {
      "package_name": "handlebars",
      "severity": "HIGH",
      "title": "Handlebars template injection",
      "installed_version": "4.7.6",
      "patched_version": "4.7.7",
      "cve": "CVE-2021-23369",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2021-23369"
    }
  ],
  "summary": {
    "high": 2,
    "medium": 1,
    "low": 0
  }
}

Example Output (pip vulnerabilities):

{
  "target": "/home/user/projects/data-pipeline",
  "package_manager": "pip",
  "total_vulnerabilities": 2,
  "vulnerabilities": [
    {
      "package_name": "requests",
      "severity": "HIGH",
      "title": "CVE-2023-32681",
      "installed_version": "2.27.1",
      "patched_version": "2.28.1",
      "cve": "CVE-2023-32681",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2023-32681"
    },
    {
      "package_name": "urllib3",
      "severity": "MEDIUM",
      "title": "CVE-2020-26137",
      "installed_version": "1.26.5",
      "patched_version": "1.26.16",
      "cve": "CVE-2020-26137",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2020-26137"
    }
  ],
  "summary": {}
}

Example Output (cargo vulnerabilities):

{
  "target": "/home/user/projects/rust-service",
  "package_manager": "cargo",
  "total_vulnerabilities": 1,
  "vulnerabilities": [
    {
      "package_name": "openssl",
      "severity": "HIGH",
      "title": "Use-after-free in OpenSSL",
      "installed_version": "0.10.35",
      "patched_version": "0.10.38",
      "cve": "CVE-2022-3358",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2022-3358"
    }
  ],
  "summary": {}
}

secret_scan

Scans source code for hardcoded secrets, credentials, and API keys. Uses gitleaks or trufflehog for pattern-based detection.

Input Schema:

{
  "target": "string (required)",
  "max_results": "integer (default: 100, range: 1-500)"
}

Parameters:

• target: Absolute path to project directory to scan
• max_results: Maximum number of secret findings to return (default: 100)

Detection Patterns (varies by tool):

• AWS credentials and API keys
• GitHub tokens and personal access tokens
• Database connection strings
• Private encryption keys
• Slack webhooks
• Stripe and payment API keys
• Generic passwords and secrets

Example Request:

{
  "target": "/home/user/projects/mobile-app",
  "max_results": 50
}

Example Output (gitleaks):

{
  "target": "/home/user/projects/mobile-app",
  "tool": "gitleaks",
  "total_secrets": 4,
  "secrets": [
    {
      "rule_id": "AWS API Key",
      "file": "/home/user/projects/mobile-app/config/aws.js",
      "line": 3,
      "match": "AKIAIOSFODNN7EXAMPLE...",
      "entropy": 4.2
    },
    {
      "rule_id": "GitHub Token",
      "file": "/home/user/projects/mobile-app/.env",
      "line": 12,
      "match": "ghp_1234567890abcdefghijklmnopqrstuv...",
      "entropy": 5.1
    },
    {
      "rule_id": "Slack Webhook URL",
      "file": "/home/user/projects/mobile-app/src/notifications.ts",
      "line": 45,
      "match": "https://hooks.slack.com/services/T00000000/B00000000...",
      "entropy": 3.8
    },
    {
      "rule_id": "Private Key",
      "file": "/home/user/projects/mobile-app/keys/id_rsa",
      "line": 1,
      "match": "-----BEGIN RSA PRIVATE KEY-----...",
      "entropy": 4.9
    }
  ]
}

Example Output (trufflehog):

{
  "target": "/home/user/projects/mobile-app",
  "tool": "trufflehog",
  "total_secrets": 2,
  "secrets": [
    {
      "rule_id": "Stripe API Key",
      "file": "/home/user/projects/mobile-app/config/stripe.py",
      "line": 8,
      "match": "sk_live_4eC39HqLyjWDarhtT657j8Wp...",
      "entropy": null
    },
    {
      "rule_id": "Database Connection String",
      "file": "/home/user/projects/mobile-app/database.js",
      "line": 2,
      "match": "postgresql://user:password@localhost:5432/db...",
      "entropy": null
    }
  ]
}

sbom_generate

Generates a Software Bill of Materials (SBOM) documenting all software components, versions, and licenses. Supports CycloneDX and SPDX formats.

Input Schema:

{
  "target": "string (required)",
  "format": "string (default: 'cyclonedx')"
}

Parameters:

• target: Absolute path to project directory
• format: SBOM output format. Options:
  • cyclonedx (default) - CycloneDX JSON format (OWASP standard)
  • spdx - SPDX JSON format (Linux Foundation standard)

Example Request:

{
  "target": "/home/user/projects/warehouse",
  "format": "cyclonedx"
}

Example Output (CycloneDX):

{
  "target": "/home/user/projects/warehouse",
  "format": "cyclonedx",
  "total_components": 47,
  "components": [
    {
      "type": "library",
      "name": "fastapi",
      "version": "0.104.1",
      "purl": "pkg:pypi/[email protected]",
      "license": "MIT"
    },
    {
      "type": "library",
      "name": "sqlalchemy",
      "version": "2.0.23",
      "purl": "pkg:pypi/[email protected]",
      "license": "MIT"
    },
    {
      "type": "library",
      "name": "pydantic",
      "version": "2.5.0",
      "purl": "pkg:pypi/[email protected]",
      "license": "MIT"
    },
    {
      "type": "library",
      "name": "requests",
      "version": "2.31.0",
      "purl": "pkg:pypi/[email protected]",
      "license": "Apache-2.0"
    },
    {
      "type": "library",
      "name": "redis",
      "version": "5.0.1",
      "purl": "pkg:pypi/[email protected]",
      "license": "MIT"
    }
  ]
}

Example Output (SPDX):

{
  "target": "/home/user/projects/warehouse",
  "format": "spdx",
  "total_components": 47,
  "components": [
    {
      "type": "library",
      "name": "express",
      "version": "4.18.2",
      "purl": "pkg:npm/[email protected]",
      "license": "MIT"
    },
    {
      "type": "library",
      "name": "mongoose",
      "version": "7.6.3",
      "purl": "pkg:npm/[email protected]",
      "license": "Apache-2.0"
    },
    {
      "type": "library",
      "name": "dotenv",
      "version": "16.3.1",
      "purl": "pkg:npm/[email protected]",
      "license": "BSD-2-Clause"
    }
  ]
}

Prerequisites

System Requirements

• Node.js 18+ or Bun runtime
• Unix-like system (macOS, Linux) or WSL on Windows

Required External Tools

The server delegates to command-line security tools. Install before use:

Semgrep (SAST)

# macOS
brew install semgrep

# Linux / Python (any OS)
pip install semgrep

# Verify installation
semgrep --version

Dependency Auditors

# npm (built-in with Node.js)
npm --version

# Python
pip install pip-audit
pip-audit --version

# Rust
cargo install cargo-audit
cargo audit --version

Secret Scanning (choose one or both)

# gitleaks (recommended)
brew install gitleaks
gitleaks version

# OR trufflehog (alternative)
brew install trufflehog
trufflehog version

SBOM Generation

npm install -g @cyclonedx/cdxgen
cdxgen --version

Verification Script

Verify all dependencies are installed:

#!/bin/bash
echo "Checking required tools..."
command -v semgrep >/dev/null 2>&1 && echo "✓ semgrep" || echo "✗ semgrep"
command -v pip-audit >/dev/null 2>&1 && echo "✓ pip-audit" || echo "✗ pip-audit"
command -v cargo >/dev/null 2>&1 && echo "✓ cargo-audit" || echo "✗ cargo-audit (install: cargo install cargo-audit)"
command -v gitleaks >/dev/null 2>&1 && echo "✓ gitleaks" || echo "✗ gitleaks (or trufflehog)"
command -v cdxgen >/dev/null 2>&1 && echo "✓ cdxgen" || echo "✗ cdxgen"

Installation

Option 1: Using Bun (Recommended)

# Clone the repository
git clone https://github.com/your-org/mcp-servers.git
cd mcp-servers/code-audit

# Install dependencies
bun install

# Build the server
bun run build

# Start the server (stdio mode)
bun run src/index.ts

Option 2: Using npm

# Install dependencies
npm install

# Build (requires build script in package.json)
npm run build

# Start
node dist/index.js

Build Output

The build process generates:

• dist/index.js - Compiled server entry point
• dist/tools/*.js - Compiled tool modules
• dist/*.js - All compiled TypeScript files

Usage

Stdio Transport (Default)

The server uses stdio (stdin/stdout) for MCP communication. The server reads requests from stdin and writes responses to stdout.

# Start the server in your terminal
bun run src/index.ts

# In another terminal, send requests via stdio
# (handled by MCP client implementations)

Claude Desktop Configuration

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

{
  "mcpServers": {
    "code-audit": {
      "command": "bun",
      "args": ["/absolute/path/to/code-audit/src/index.ts"],
      "env": {}
    }
  }
}

Note: Use absolute path to the index.ts file. Bun will handle compilation automatically.

Claude Code MCP Settings

In your Claude Code IDE settings, add the MCP configuration:

{
  "mcpServers": {
    "code-audit": {
      "command": "bun",
      "args": ["/Users/username/Documents/code/mcp-servers/code-audit/src/index.ts"]
    }
  }
}

Environment Variables

Optional environment variables for tool configuration:

# Semgrep registry endpoint (if using custom registry)
SEMGREP_REGISTRY_ENDPOINT="https://registry.semgrep.dev"

# Gitleaks configuration
GITLEAKS_CONFIG="/path/to/gitleaks.toml"

# Tool timeout (milliseconds, default 120000)
TOOL_TIMEOUT_MS="120000"

Example Usage in Claude

Once configured, you can ask Claude to:

"Audit the security of the code in /home/user/projects/api-server"

Claude will:
1. Run semgrep_scan to find SAST issues
2. Run dep_audit to check dependencies
3. Run secret_scan for hardcoded credentials
4. Run sbom_generate to document components
5. Synthesize findings into a security report

Security Considerations

Path Validation

All directory paths are validated before scanning:

1. Resolution: Paths are normalized and resolved to absolute form

2. Blocking: Access to sensitive system directories is blocked:

   • /etc - System configuration
   • /proc - Process information
   • /sys - System kernel interface
   • /dev - Device files
   • /var/run - Runtime state

3. Existence Check: Only existing directories are scanned

4. Type Check: Target must be a directory, not a file

Error Examples:

{
  "content": [
    {
      "type": "text",
      "text": "Error: Access to /etc is not allowed"
    }
  ],
  "isError": true
}

Timeout Protection

Each tool invocation has timeout protection:

• Default Timeout: 120 seconds (120,000 ms)
• Configurable: Via optional env variable
• Process Termination: Killed automatically on timeout

Error Example:

{
  "content": [
    {
      "type": "text",
      "text": "Error: Command timed out after 120000ms"
    }
  ],
  "isError": true
}

Tool Execution Safety

• execFile API: Uses Node.js execFile (not shell exec), preventing command injection
• Max Buffer: 20 MB output limit per command
• Error Handling: Non-zero exit codes captured as errors
• No Shell Interpolation: Tool arguments passed as array elements

Recommended Practices

1. Limit Directory Scope: Scan only necessary directories (not entire filesystem)
2. Monitor Timeout: Increase timeout for large repositories if needed
3. Review Credentials: Verify secret_scan findings manually before remediation
4. Dependency Updates: Keep semgrep, gitleaks, and audit tools current
5. Configuration Hardening: Use restricted-access rules in Semgrep for high-security contexts

Performance Notes

Scanning Time Estimates

| Tool | Small Project (< 1K files) | Medium Project (1-10K files) | Large Project (> 10K files) | |------|---------------------------|------------------------------|----------------------------| | semgrep_scan | 10-30s | 30-60s | 60-120s | | dep_audit | 5-10s | 10-20s | 20-30s | | secret_scan | 5-15s | 15-40s | 40-120s | | sbom_generate | 5-20s | 20-50s | 50-120s |

Optimization Tips

1. Exclude Directories: Use .semgrep.yml to exclude large test/vendor directories
2. Parallel Execution: Run multiple tools concurrently in Claude
3. Incremental Scanning: Scan changed files only when possible
4. Result Limiting: Use max_findings and max_results parameters

Troubleshooting

"Tool not found" Errors

Semgrep:

# Error: "semgrep not found"
pip install semgrep
semgrep --version

gitleaks:

# Error: "gitleaks not found"
brew install gitleaks
gitleaks version

cdxgen:

# Error: "cdxgen not found"
npm install -g @cyclonedx/cdxgen
cdxgen --version

Timeout Issues

Increase timeout for large repositories:

# Set environment variable (Linux/macOS)
export TOOL_TIMEOUT_MS="300000"  # 5 minutes
bun run src/index.ts

Parse Errors

If tools output non-JSON results:

1. Verify tool version compatibility
2. Check tool output manually:

   cd /path/to/project
   semgrep scan --config auto --json .

3. Report issue with tool output format

Empty Dependency Audit Results

If dep_audit returns no vulnerabilities:

1. Verify correct project structure:

   • npm: Has package-lock.json or package.json
   • pip: Has requirements.txt, setup.py, or pyproject.toml
   • cargo: Has Cargo.lock or Cargo.toml

2. Run audit manually to debug:

   cd /path/to/project
   npm audit --json

API Reference

Tool Input/Output Format

All tools accept Zod-validated JSON schemas and return structured JSON responses.

Request Format (internal MCP protocol):

{
  "name": "tool_name",
  "arguments": {
    "target": "/path/to/project",
    "config": "auto"
  }
}

Response Format (all tools):

{
  "content": [
    {
      "type": "text",
      "text": "{...JSON result...}"
    }
  ],
  "isError": false
}

Error Responses

On error, the response includes isError: true:

{
  "content": [
    {
      "type": "text",
      "text": "Error: Invalid target directory"
    }
  ],
  "isError": true
}

License

MIT

Contributing

To contribute improvements to the Code Audit MCP server:

1. Fork the repository
2. Create a feature branch
3. Make changes and test thoroughly
4. Submit a pull request with clear description

Testing Requirements:

• Test all four tools with sample projects
• Verify timeout and error handling
• Confirm path validation works correctly

Related Resources

• Model Context Protocol Specification
• Semgrep Documentation
• gitleaks GitHub Repository
• trufflehog GitHub Repository
• CycloneDX SBOM Standard
• SPDX SBOM Standard
• OWASP Top 10