# joergmichno/clawguard-mcp [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/joergmichno/clawguard-mcp  
**GitHub Stars:** 1  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/joergmichno-clawguard-mcp

## Description
Security scanner for AI agents that detects prompt injections using 42+ regex patterns

## Tools
Capabilities this server exposes over MCP:

- **scan_text** — Scan a single text for prompt injection threats
- **scan_batch** — Scan up to 10 texts in one call
- **get_patterns** — List all 225 detection patterns by category
- **get_usage** — Check your API usage and remaining quota
- **health_check** — Verify the Shield API is running

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

```json
"mcpServers": {
  "clawguard-mcp": {
    "command": "uvx",
    "args": ["clawguard-mcp"],
    "env": {
      "CLAWGUARD_API_KEY": "",
      "CLAWGUARD_BASE_URL": ""
    }
  }
}
```

**Requires environment variables:** `CLAWGUARD_API_KEY`, `CLAWGUARD_BASE_URL` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What joergmichno/clawguard-mcp does

ClawGuard MCP makes ClawGuard Shield’s prompt injection scanner available through the Model Context Protocol. It is intended for use from Claude Desktop, Claude Code, Cursor, or another MCP-compatible client. Scan results can identify whether text is clean, provide a risk score and severity, and return matched findings such as instruction overrides.

The Shield API evaluates content against 225 detection patterns. Pattern categories include prompt injection, jailbreak attempts, data exfiltration, social engineering, and encoding attacks. The scanner supports 15 languages and maps individual patterns to OWASP LLM and Agentic Top 10 areas. Detection uses a deterministic regular-expression engine rather than an LLM round trip.

## How it works

The MCP process receives tool calls from the connected client and sends scanning or account requests to ClawGuard Shield. `scan_text` handles one text value, while `scan_batch` accepts up to 10 texts in a single call. The server also exposes pattern information, quota status, and an API health check.

A typical request asks the client to scan text for prompt injection. The returned data may include an `is_clean` result, a numeric `risk_score`, a severity level, and findings with the matched pattern, category, severity, and text.

## Setup and configuration

The package can be launched with `uvx clawguard-mcp`. Configure it as an MCP server and provide the required `CLAWGUARD_API_KEY`, which uses the `cgs_` key format. The default Shield endpoint is `https://prompttools.co/api/v1`; set `CLAWGUARD_BASE_URL` when using another endpoint, including a self-hosted setup.

Claude Desktop configuration uses the `command` value `uvx`, the argument `clawguard-mcp`, and the API key in the server’s environment. Claude Code can add the server with `claude mcp add clawguard -- uvx clawguard-mcp`, followed by setting the key in the environment. A free Shield tier includes 100 scans per day.

For local development, the repository uses `uv sync` for installation and `uv run pytest` for tests. MCP Inspector can launch the local package for manual testing.

## Tools and capabilities

- `scan_text`: inspect one text for prompt injection threats.
- `scan_batch`: inspect up to 10 texts in one request.
- `get_patterns`: list the 225 detection patterns and their categories.
- `get_usage`: view API consumption and remaining quota.
- `health_check`: verify that the Shield API is available.

## Limitations and notes

Scanning depends on access to the Shield API and a valid API key. The free tier limits usage to 100 scans per day. The repository is retained as a historical reference and identifies ModVet as its successor for pre-install scanning of Claude Code mods; ModVet is described as offline and deterministic, without an LLM.

_Full upstream README: https://allmcps.com/mcp/joergmichno-clawguard-mcp/readme_