DevGuard-MCP
# 🛡️ DevGuard-MCP
[](https://github.com/YoruichiYams/devguard-mcp/actions/workflows/ci.yml)
[](https://pypi.org/project/devguard-mcp/)
[](https://pypi.org/project/devguard-mcp/)
[](https://opensource.org/licenses/MIT)
**DevGuard-MCP** is a fast, local, deterministic developer workspace auditor, safe cleaner, advanced security secret scanner, and Model Context Protocol (MCP) server.
---
## ✨ Features
- **🔍 Workspace Auditing**: Detects reclaimable build artifacts, dependencies, and caches (`node_modules`, `.venv`, `.pytest_cache`, `__pycache__`, `target/`, `build/`, `dist/`, `.next/`, logs).
- **🔒 Advanced Security Engine**:
- **High-Confidence Signatures**: Real-time regex pattern matching for AWS Access Keys (`AKIA...`), GitHub Tokens (`ghp_...`, `github_pat_...`), OpenAI API keys (`sk-...`), and Private RSA/SSH/EC keys.
- **Shannon Information Entropy ($H$)**: Flags suspicious high-entropy pseudo-random tokens and passwords in configuration files.
- **Automated Secret Masking**: Obfuscates sensitive credentials in logs and outputs (e.g. `sk-proj-abc1...****`).
- **📜 SQLite History & Delta Tracking**: Automatically logs workspace scans and cleanup events to a local SQLite database (`~/.devguard/history.db`) to track workspace hygiene changes over time.
- **🗑️ Safe Cleanup**: Built-in safety mechanisms with dry-run previews, OS Recycle Bin / Trash quarantine (`send2trash`), and protected system path verification.
- **📊 Rich CLI**: Beautiful terminal user interface powered by Typer and Rich (tables, progress bars, colored summaries, JSON outputs).
- **🤖 MCP Server**: Native Model Context Protocol support to let AI assistants (Claude, Antigravity, Cursor, etc.) safely audit workspaces and diagnose security hygiene.
---
## 🚀 Quick Start
### Installation
Install via pip or run directly using uvx:
```bash
# Using pip
pip install devguard-mcp
# Or run ephemerally with uvx
uvx devguard-mcp --help
```
### CLI Usage
```bash
# 🔍 Scan workspace for reclaimable artifacts
devguard scan
# 🔒 Run deep security audit (detects .env exposures and leaked API keys)
devguard audit
# 📜 View historical audit timeline and metric deltas
devguard history --limit 10
# 🗑️ Preview safe cleanup (dry-run simulation by default)
devguard clean --dry-run
# 🗑️ Clean specific artifacts into OS Recycle Bin
devguard clean --type node_modules --trash --execute
```
---
## 🤖 MCP Server Configuration
Configure DevGuard with your MCP client (Claude Desktop, Antigravity IDE, Cursor):
### Example `mcp_config.json`:
```json
{
"mcpServers": {
"devguard": {
"command": "uvx",
"args": ["devguard-mcp"]
}
}
}
```
### Available MCP Tools
- `scan_workspace(path, max_depth, artifact_types, min_size_mb)`: Discover cleanable artifacts and disk usage.
- `audit_workspace_secrets(path)`: Deep security audit detecting exposed credentials, `.env` files, and health scores.
- `get_audit_history(path, limit)`: Query historical scan reports and cleanup logs.
- `preview_cleanup(path, target_paths, artifact_types, older_than_days)`: Dry-run simulation of safe cleanup.
- `clean_workspace(path, target_paths, artifact_types, mode, dry_run)`: Clean artifacts with safety boundaries.
- `get_workspace_health(path)`: High-level overview of disk footprint and security health.
---
## 📄 License
MIT License © 2026 DevGuard Team
TDQS
Scored across 6 tools
Most tools have clearly distinct purposes: audit focuses on security risks, scan on reclaimable artifacts, preview/clean are paired, history and health are distinct. The only potential overlap is between audit_workspace_secrets and scan_workspace, as both may detect secrets, but their descriptions emphasize different objectives (security audit vs artifact discovery).
All tool names follow a consistent verb_noun pattern in snake_case (audit, preview, get, clean, scan), with only minor variation like audit_workspace_secrets adding an extra noun. The style is uniform and predictable, making it easy to infer behavior from names.
Six tools is well-scoped for a workspace hygiene server, covering scanning, auditing, previewing, cleaning, health metrics, and history. Each tool serves a distinct function without redundancy, and the count feels neither sparse nor bloated.
The surface covers the core lifecycle: scan (discover), audit (security), preview (simulate), clean (execute), plus health and history. Minor gaps exist such as no explicit tool for managing exclusion lists or restoring from recycle bin, but these are not critical for the primary workflow.