ZAP-MCP: Model Context Protocol for OWASP ZAP
About
ZAP-MCP is a server that bridges AI models (like Claude) with OWASP ZAP through the Model Context Protocol (MCP). It enables AI-driven security testing by exposing standardized functions that AI models can call to perform scans, get alerts, and generate reports.
Details
- Author
- ajtazer
- GitHub stars
- 20
- Downloads
- 474
- Categories
- Other, AI, Security
Jump to
- AI-driven security testing through LLM interaction
- Real-time scan monitoring with instant alerts
- Automated security report and recommendation generation
- Flexible integration with various MCP-compatible AI models
- WebSocket-based real-time updates and communication
Install dependencies with pip install -r requirements.txt, run ./setup_mcp.sh, and configure claude_desktop_config.json with your ZAP API key and URL. Start the MCP server with mcp-server --config claude_desktop_config.json --model-dir ./models, then configure your Claude desktop app to set the WebSocket URL to ws://localhost:7456/ws. After setup, Claude will have access to ZAP scanning tools.
ZAP-MCP: Model Context Protocol for OWASP ZAP
A powerful integration between OWASP ZAP and AI models through the Model Context Protocol (MCP). This project enables AI-driven security testing by allowing AI models to directly interact with ZAP's scanning capabilities.
Star History
Overview
ZAP-MCP provides a bridge between AI models (like Claude) and OWASP ZAP, enabling automated security testing and analysis. It uses a client-server architecture where ZAP-MCP acts as the server, exposing standardized functions that can be called by AI models through the MCP protocol.
Features
- AI-Driven Security Testing: Enable AI models to perform security scans and analysis
- Real-time Scan Monitoring: Track scan progress and get instant alerts
- Automated Analysis: Generate security reports and recommendations
- Flexible Integration: Works with various AI models through the MCP protocol
- WebSocket Communication: Real-time updates and interactions
Prerequisites
- Python 3.8+
- OWASP ZAP running locally or remotely
- Claude Desktop App (or other MCP-compatible client)
Installation
1. Clone the repository:
git clone https://github.com/tazer/ZAP-MCP.git
cd ZAP-MCP
2. Install dependencies:
pip install -r requirements.txt
3. Set up the MCP server:
./setup_mcp.sh
4. Configure ZAP-MCP:
- Copy claude_desktop_config.json to your Claude app's config directory
- Update the ZAP API key and URL in the config file
Usage
1. Start the MCP server:
mcp-server --config claude_desktop_config.json --model-dir ./models
2. Configure your Claude desktop app:
- Open Claude app settings
- Enable MCP server integration
- Set the WebSocket URL to: ws://localhost:7456/ws
3. Start using ZAP-MCP:
- The Claude app will now have access to ZAP scanning tools
- You can request security scans, get alerts, and generate reports
Available Tools
The MCP server exposes these ZAP-specific tools:
- start_scan: Start a new ZAP scan on a target URL
- get_scan_status: Check the status of a running scan
- get_alerts: Get all alerts from the current scan
- get_scan_summary: Get a summary of the current scan
Configuration
The claude_desktop_config.json file contains all necessary settings:
{
"mcp_server": {
"host": "localhost",
"port": 7456,
"model": "claude-instant-v1",
"max_tokens": 1000,
"temperature": 0.7,
"zap_api_key": "your-zap-api-key",
"zap_url": "http://localhost:8080"
},
"local_models": {
"path": "./models",
"prefer_local": true
},
"zap_settings": {
"scan_timeout": 300,
"max_concurrent_scans": 5,
"alert_threshold": "HIGH",
"scan_policy": "default"
}
}
Development
This project was developed using Cursor AI, an intelligent coding assistant that helped streamline the development process and ensure code quality.
Author
- TAZER - Initial work and maintenance
License
This project is licensed under the MIT License - see the LICENSE file for details.
Acknowledgments
- OWASP ZAP team for their excellent security testing tool
- Anthropic for Claude AI
- Cursor AI for their assistance in development
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.




