# GUCCI-atlasv/skillssafe-mcp [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/GUCCI-atlasv/skillssafe-mcp  
**GitHub Stars:** 1  
**npm Downloads (last month):** 189  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/gucci-atlasv-skillssafe-mcp

## Description
Free AI agent skill security scanner. Scan SKILL.md, MCP configs, and system prompts for credential theft, prompt injection, zero-width character attacks, and ClawHavoc indicators. Supports OpenClaw, Claude Code, Cursor, and Codex. No signup required.

## Tools
Capabilities this server exposes over MCP:

- **scan_skill** — Scan an AI agent skill file for security threats before installation.
- **get_report** — Retrieve a previously generated scan report.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "skillssafe-mcp": {
    "command": "npx",
    "args": ["-y","skillssafe-mcp"]
  }
}
```

## Documentation

## What GUCCI-atlasv/skillssafe-mcp does

GUCCI-atlasv/skillssafe-mcp MCP server exposes SkillsSafe security scanning to AI agent clients. It is designed for reviewing SKILL.md files, MCP configurations, and system prompts before they are used or installed. The scanner looks for credential theft, data exfiltration, prompt injection, hidden zero-width characters, ClawHavoc malware indicators, shell injection, reverse shells, excessive scope, and memory poisoning.

The primary workflow starts with `scan_skill`. You can provide either a URL, such as a GitHub raw or ClawHub URL, or the skill text directly. The result includes an `INSTALL`, `REVIEW`, or `BLOCK` decision, a risk score from 0 to 100, and detected threats with severity information. Each scan also returns a scan ID for later retrieval.

## How it works

A client calls `scan_skill` with the material to assess. The `lang` parameter controls the response language and accepts `en`, `zh`, or `ja`; English is the default. The server passes the supplied URL or content to SkillsSafe and returns the structured scan result.

Use `get_report` with the scan ID from a completed scan when the client needs the previously generated report. This separates the initial scan decision from access to the full report and avoids resubmitting the same material.

GUCCI-atlasv/skillssafe-mcp MCP server can be started as a local stdio process with `npx -y skillssafe-mcp`. For clients that support Server-Sent Events, the documented remote endpoint is `https://mcp.skillssafe.com/sse`.

## Setup and configuration

Claude Desktop can register the server in its MCP configuration by invoking `npx` with the package name and the `-y` argument. Cursor uses the same command and arguments in `.cursor/mcp.json`. These configurations run the package through Node's package runner rather than requiring a separate local installation step.

The server does not require an API key, account, or signup according to the project documentation. No environment variables are listed as required. Remote use is available through the documented SSE URL for compatible clients.

The project is listed in the official MCP Registry as `com.skillssafe/scanner`. It is also referenced in Smithery and Glama, but those listings are directory or registry references rather than additional configuration methods described here.

## Tools and capabilities

- `scan_skill`: examines a skill URL or raw skill content and returns a decision, numeric risk score, threat details, and scan ID.
- `get_report`: retrieves a report using a scan ID returned by `scan_skill`.
- Threat coverage includes credential theft, exfiltration, prompt injection, zero-width attacks, ClawHavoc indicators, shell and reverse-shell behavior, scope creep, and memory poisoning.
- Response language selection is available for English, Chinese, and Japanese.

## Limitations and notes

A scan can return `REVIEW`, so users still need to assess findings that are not an automatic block.

It also does not list configurable environment variables, authentication settings, rate limits, or retention behavior for reports. The remote endpoint should therefore be used only by clients that support the documented SSE transport.

_Full upstream README: https://allmcps.com/mcp/gucci-atlasv-skillssafe-mcp/readme_