https://github.com/naufalafif/agent-guard
Native macOS menu bar app that scans MCP servers and AI agent skills for security threats
https://github.com/naufalafif/agent-guard
ai-agents cisco macos mcp menu-bar security swift
Last synced: 4 months ago
JSON representation
Native macOS menu bar app that scans MCP servers and AI agent skills for security threats
- Host: GitHub
- URL: https://github.com/naufalafif/agent-guard
- Owner: naufalafif
- Created: 2026-03-22T23:03:14.000Z (4 months ago)
- Default Branch: staging
- Last Pushed: 2026-03-23T14:29:27.000Z (4 months ago)
- Last Synced: 2026-03-23T17:54:13.302Z (4 months ago)
- Topics: ai-agents, cisco, macos, mcp, menu-bar, security, swift
- Language: Swift
- Homepage: https://github.com/naufalafif/agent-guard
- Size: 4.81 MB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
Awesome Lists containing this project
README
# AgentGuard
Native macOS menu bar app that scans your MCP servers and AI agent skills for security threats.
    
## Install
```bash
brew tap naufalafif/tap
brew install --cask agent-guard
```
Or build from source:
```bash
git clone https://github.com/naufalafif/agent-guard.git
cd agent-guard
make install
```
Requires macOS 13+. The app auto-installs scanner dependencies on first launch.
## What it scans
| Scanner | What it checks |
|---------|---------------|
| [mcp-scanner](https://github.com/cisco-ai-defense/mcp-scanner) | MCP server configs — Claude Desktop, Cursor, VS Code, Windsurf, Zed |
| [skill-scanner](https://github.com/cisco-ai-defense/skill-scanner) | Agent skill packages — Cursor rules, Claude skills, and other agent instruction files |
Powered by [Cisco AI Defense](https://github.com/cisco-ai-defense). YARA rules + static analysis. Everything runs locally.
## Usage
- **Menu bar icon** — green = all clear, red + count = findings, half shield = scanning
- **Click** the icon to see findings, grouped by MCP servers and AI skills
- **Expand** a finding to see details — threat name, category, rule
- **Mute** false positives with confirmation, unmute anytime
- **Settings** (gear icon) — scan interval, skill directories, launch at login
> **Tip:** AgentGuard scans common locations (`~/.cursor/rules`, `~/.claude/skills`, `~/.claude/plugins`, etc.) by default — including Claude Code marketplace plugins. If you keep skills inside project directories, add your workspace path in Settings.
## Uninstall
```bash
brew uninstall --cask agent-guard
rm -rf ~/.cache/mcp-scan ~/.config/mcp-scan # optional: remove data
```
## Contributing
See [CONTRIBUTING.md](CONTRIBUTING.md) for development setup, architecture, and CI/CD details.
## License
MIT