Garak MCP

A lightweight MCP server that supports the Garak LLM vulnerability scanning tool, providing functions such as listing model types, models, and attack probes, and can run attack tests to generate reports.

Security Developer tools #Vulnerability Scanning #LLM Security #MCP Protocol #Model Testing .Python

rating : 2 points

downloads : 7.6K

update time : 2025-04-29

Open Site

What is Garak MCP Server?

The Garak MCP Server provides a simple interface to test language models for security vulnerabilities. It supports multiple model types and various attack probes to identify potential weaknesses in AI systems.

How to use Garak MCP?

You can interact with the server through supported clients like Claude Desktop or Cursor. The server provides commands to list available models, probes, and run security tests.

Use Cases

Ideal for AI security researchers, developers testing their LLM implementations, and organizations evaluating model security before deployment.

Key Features

Multiple Model Support

Works with Ollama, OpenAI, HuggingFace and GGML model types

Comprehensive Probe Library

Includes various pre-built security probes from Garak framework

Lightweight Server

Minimal resource requirements for easy deployment

Advantages

Easy to set up and configure

Supports multiple popular model types

Provides standardized vulnerability reports

Limitations

Currently limited to specific model types

Reporting features need improvement

Some model integrations not fully tested

Getting Started

Install Prerequisites

Ensure you have Python 3.11+ and uv installed

Clone Repository

Get the server code from GitHub

Configure Client

Set up your MCP client (like Cursor) to connect to the server

Example Scenarios

Testing Local Ollama Model

Check a locally running Ollama model for prompt injection vulnerabilities

Listing Available Probes

Discover what security tests are available

Frequently Asked Questions

What clients support this MCP server?

How do I view the scan results?

Can I add custom probes?

Additional Resources

Garak Framework

The underlying security framework used by this server

MCP Protocol

Documentation on Model Context Protocol

Ollama

For running local models

🚀 MCP Server for Garak LLM Vulnerability Scanner

A lightweight MCP (Model Context Protocol) server suitable for Garak.

Example:

https://github.com/user-attachments/assets/f6095d26-2b79-4ef7-a889-fd6be27bbbda

🚀 Quick Start

This is a lightweight MCP (Model Context Protocol) server for Garak. It provides several useful tools to interact with different models and run vulnerability scans.

✨ Features

Overview

Property	Details
list_model_types	List all available model types (ollama, openai, huggingface, ggml)
list_models	List all available models based on the given model type
list_garak_probes	List all available Garak attacks/probes
get_report	Get the report of the last run
run_attack	Run an attack using the specified model and probe

Detailed Description

list_model_types
- List all model types available for attacks.
- Return a list of supported model types (ollama, openai, huggingface, ggml).
list_models
- List all available models based on the given model type.
- Input parameters:
  - model_type (string, required): The model type to list (ollama, openai, huggingface, ggml).
- Return a list of available models of the specified type.
list_garak_probes
- List all available Garak attacks/probes.
- Return a list of all available attacks/probes.
get_report
- Get the report of the last run.
- Return the path to the report file.
run_attack
- Run an attack using the specified model and probe.
- Input parameters:
  - model_type (string, required): The model type to use.
  - model_name (string, required): The name of the model to use.
  - probe_name (string, required): The name of the attack/probe to use.
- Return a list of discovered vulnerabilities.

📦 Installation

Prerequisites

Python 3.11 or higher: This project requires Python 3.11 or later.
```
# Check your Python version
python --version
```
Install uv: A fast Python package installer and resolver.
```
pip install uv
```
Or use Homebrew:
```
brew install uv
```
Optional: Ollama: If you want to run attacks on Ollama models, make sure the Ollama server is running.
```
ollama serve
```

Steps

Clone this repository:

git clone https://github.com/BIGdeadLock/Garak-MCP.git

Configure your MCP host (e.g., Claude Desktop, Cursor, etc.):

{
  "mcpServers": {
    "garak-mcp": {
      "command": "uv",
      "args": ["--directory", "path-to/Garak-MCP", "run", "garak-server"],
      "env": {}
    }
  }
}