🛡️ CLAUDIT-SEC

Security audit tool for Claude Desktop on macOS and Windows — including CoWork, extensions, plugins, MCP servers, connectors, and scheduled tasks.

One command. Full visibility. Read-only.

⚠️ Windows support is a work in progress. We're aware of a few kinks and bugs and wanted to get something out sooner rather than later. Community feedback and contributions are welcome.

🤔 Why

Claude Desktop introduces a new class of endpoint risk: AI agents with autonomous execution, persistent scheduled tasks, MCP server integrations, browser-control extensions, and OAuth-authenticated connectors to external services. Most of this configuration lives in JSON files scattered across multiple directories with no centralised visibility.

CLAUDIT gives you that visibility in a single command.

📝 A note on "Code": Claude Desktop includes a built-in agent coding feature called Code (visible in the app's sidebar). This is not the same as Claude Code, the standalone terminal CLI. CLAUDIT primarily audits Claude Desktop and its CoWork features. It does include a basic check of the Claude Code settings file (~/.claude/settings.json on macOS, %USERPROFILE%\.claude\settings.json on Windows), but the focus is squarely on the Desktop app.

📋 What It Audits

Area	What's Checked
🖥️ Desktop Settings	keepAwakeEnabled, sidebar/menuBar preferences
🤖 CoWork Settings	Scheduled tasks, web search, browser use, dispatch (mobile→desktop), network mode, egress policy, enabled plugins, marketplaces
🏢 Workspaces	Multi-workspace detection, account names, session counts, org indicators (DXT-managed, org-plugins, dispatch-bridge)
🔌 MCP Servers	Server names, commands, arguments, environment variable keys
🧩 Extensions (DXT)	Installed extensions, signature status, dangerous tool grants
⚙️ Extension Settings	Per-extension allowed directories and configuration
🚦 Extension Governance	Allowlist enabled/disabled, blocklist entries
📦 Plugins	Installed, remote (org-deployed), cached (downloaded)
🪝 Plugin Hooks	Lifecycle hooks executing shell commands (PreToolUse, PostToolUse, Stop, etc.)
🔗 Connectors	OAuth-authenticated web services, desktop integrations
🎯 Skills	User-created, scheduled, session-local, and plugin skills across 9 paths
⏰ Scheduled Tasks	Task names, cron expressions (with plain English translation)
🔐 App Config	Network mode, extension allowlist/blocklist keys, device identifiers
📲 Dispatch	Bridge state (OFF/CONFIGURED/ON), active session detection via hostLoopMode and bridge-state.json
🔇 Disabled MCP Tools	Per-session tools explicitly disabled (with dangerous tool callout)
🏃 Runtime State	Running processes, sleep assertions, LaunchAgents, crontab entries
🍪 Cookies	Cookies and Cookies-journal presence

📖 For a detailed breakdown of every individual check, what it means, and why it matters, see the Findings Reference.

⚡ Getting Started

Prerequisites

macOS:

Requirement	How to check	How to install
🍎 macOS	You're on a Mac	—
🐚 zsh	zsh --version	Ships with macOS since Catalina
🔧 jq	jq --version	brew install jq

Windows:

Requirement	How to check	How to install
🪟 Windows 10/11	You're on a PC	—
⚡ PowerShell 5.1+	$PSVersionTable.PSVersion	Ships with Windows 10+
—	No additional dependencies	Fully self-contained

Install & Run

macOS:

git clone https://github.com/HarmonicSecurity/claudit-sec.git
cd claudit-sec
chmod +x claude_audit.sh
./claude_audit.sh

Windows:

git clone https://github.com/HarmonicSecurity/claudit-sec.git
cd claudit-sec
powershell -ExecutionPolicy Bypass -File claude_audit.ps1

That's it. The script reads your Claude configuration and prints a colour-coded report to the terminal. It never modifies anything.

🎛️ Usage

macOS:

./claude_audit.sh [OPTIONS]

Options:
  --html [FILE]    Generate a standalone HTML report
  --json           Output structured JSON
  --user USER      Audit a specific user
  --all-users      Audit all users with Claude data (requires root)
  -q, --quiet      Only show WARN and CRITICAL findings
  --version        Print version and exit
  -h, --help       Show usage

Windows:

powershell -ExecutionPolicy Bypass -File claude_audit.ps1 [OPTIONS]

Options:
  -Html [FILE]     Generate a standalone HTML report
  -Json            Output structured JSON
  -User USER       Audit a specific user
  -AllUsers        Audit all users with Claude data (requires admin)
  -Quiet           Only show WARN and CRITICAL findings
  -Version         Print version and exit
  -Help            Show usage

Examples

macOS:

# Default: colour output in terminal
./claude_audit.sh

# Only warnings and critical findings
./claude_audit.sh -q

# Standalone HTML report
./claude_audit.sh --html

# JSON for SIEM ingestion
./claude_audit.sh --json > audit.json

# Specific user
./claude_audit.sh --user jsmith

# All users (run as root via MDM — FleetDM, Jamf, Mosyle, CrowdStrike RTR)
sudo ./claude_audit.sh

Windows:

# Default: colour output in terminal
powershell -ExecutionPolicy Bypass -File claude_audit.ps1

# Only warnings and critical findings
powershell -ExecutionPolicy Bypass -File claude_audit.ps1 -Quiet

# Standalone HTML report
powershell -ExecutionPolicy Bypass -File claude_audit.ps1 -Html

# JSON for SIEM ingestion
powershell -ExecutionPolicy Bypass -File claude_audit.ps1 -Json > audit.json

# Specific user
powershell -ExecutionPolicy Bypass -File claude_audit.ps1 -User jsmith

# All users (run as admin via MDM — Intune, CrowdStrike RTR)
powershell -ExecutionPolicy Bypass -File claude_audit.ps1 -AllUsers

💡 macOS: When run as root (uid 0), the script automatically discovers and scans all users with Claude data. No flags needed.

💡 Windows: When run as Administrator, the script can scan all users with the -AllUsers flag. MDM tools like Intune and CrowdStrike RTR execute scripts with elevated privileges.

📊 Output Formats

🖥️ Terminal (default)

Colour-coded output with Unicode tables and severity indicators.

🌐 HTML (--html)

Standalone dark-themed report with collapsible sections. Created with restrictive file permissions (0600).

📄 JSON (--json)

Structured output for SIEM ingestion. Sensitive fields (OAuth tokens, API keys, secrets) are automatically redacted. Multi-user scans produce a JSON array.

🚨 Severity Levels

Severity	Meaning
🟠 WARN	Increases risk surface — e.g. unsigned extensions, autonomous execution enabled
🟡 REVIEW	Needs human judgement — e.g. org-deployed plugins, MCP servers present
🔵 INFO	Informational — e.g. Claude is running, permissions granted

📖 Documentation

Doc	Description
Findings Reference	Every individual check CLAUDIT performs, what it means, why it matters (risk, compliance, AI enablement), and what to do about it

🔒 Security Properties

• Read-only — never writes to, modifies, or deletes any audited file
• No network access — all data collected from local filesystem and system commands
• Sensitive data redacted — tokens, keys, and secrets replaced with [REDACTED] in all output formats
• Minimal privileges — runs as current user; root only needed for multi-user scans
• Single file — macOS requires jq; Windows is fully self-contained (no external dependencies)
• Auditable — the entire tool is one readable script (claude_audit.sh on macOS, claude_audit.ps1 on Windows)

💜 Built with Claude Code

This project is built and maintained using Claude Code. We love it. Seriously. If you're building developer tools and haven't tried it yet, you're missing out.

📄 License

Apache License 2.0 — see LICENSE for details.