Security Scanner MCP Server

How to use

Connect your MCP client to the Security Scanner server to start a scan. Use an MCP client such as Cherry Studio, Kiro, Cursor, or Trae to point at the scanner’s HTTP endpoint. The scanner operates in two modes: a Fast mode for quick concurrent checks and a Full mode that uses comprehensive guidance from an LLM to perform deeper analysis. You can start a scan on a target MCP, monitor progress in real time, and retrieve a detailed report that includes proof of concepts and recommended fixes.

When you initiate a scan, you can reference the scanner’s HTTP URL to begin immediately, or you can use a secure token if your environment requires authentication. You will see a running progress indicator and, once complete, a full results set containing discovered tools, executed attacks, and vulnerabilities organized by severity.

To view results, request the scan details from the client interface after the scan completes. The report includes a summary with metrics such as tools discovered, injectable tools, attacks executed, and vulnerabilities found, along with detailed vulnerability entries and remediation recommendations.

How to install

pip install -r requirements.txt
# Recommended: use a virtual environment
uv init MCP  
cd MCP
echo "3.13" > .python-version
uv venv
source .venv/bin/activate
uv add "mcp[cli]" httpx

Prerequisites you need before installation: Python 3.10 or newer, the LangGraph framework, FastMCP, httpx, and MCP dependencies. Ensure you have a compatible runtime of your choice (Python/uv) installed on your system.

Additional configuration and usage notes

Start the MCP Security Scanner server with one of the following commands. The server exposes its status and endpoints at a default address you can use from your MCP client.

# Start with Python directly
python main.py
# Or use uv for a managed environment
uv run main.py

By default, the server runs at http://0.0.0.0:8000/sse, which you will reference from your MCP client configuration.

Connection options and environment

You can connect to the scanner either over HTTP or via local stdio. If you have a remote scanner endpoint, you can configure the client to point at the HTTP URL. A local, in-process run uses the Python interpreter or uvx runner.

Example configurations

{
  "mcpServers": {
    "Security-Scanner": {
      "url": "http://10.0.xx.xx:8000/sse",
      "type": "http",
      "args": []
    }
  }
}

If you prefer a local stdio run, you can start the server with Python or uv and connect to it via the local endpoint as shown above.

Output and reporting

The tool generates a comprehensive security report that includes a status, target, risk level, a summary with counts of discovered tools, injectable tools, attacks executed, and vulnerabilities found, plus detailed vulnerability entries grouped by severity. The report includes remediation suggestions and proof-of-concept details to help you remediate issues quickly.

Notes on modes and capabilities

Two scan modes are available: fast for rapid parallel checks and full for a complete LLM-guided exploration. The scanner supports multiple attack types such as command injection, code execution, SQL injection, SSRF, path traversal, IDOR, excessive data exposure, sensitive data disclosure, and hard-coded credentials, along with payload generation and asynchronous task management.

Troubleshooting and tips

If you encounter connectivity issues, verify that the server is reachable at the configured URL and that the client has permission to access the endpoint. For long-running scans, use the asynchronous task management features to monitor progress. If you see authentication requirements, supply the appropriate Authorization header with a Bearer token as needed.

Security and ethics

Use this scanner only on systems you are authorized to test. It is designed to help you identify and remedy security risks in MCP services and should be used in accordance with your organization’s security policy and applicable laws.

Automated security scanning

Automates discovery and testing of MCP service exposure for security gaps across multiple attack surfaces.

Command Injection testing

Detects execution of system commands within MCP endpoints.

Code Execution testing

Checks for arbitrary code execution vulnerabilities within MCP components.

SQL Injection testing

Evaluates database query handling to identify injection flaws.

SSRF testing

Tests access to internal resources via server-side requests.

Path Traversal testing

Assesses filesystem access via crafted paths.

IDOR testing

Verifies proper access controls to prevent unauthorized data access.

Excessive Data Exposure detection

Identifies responses that leak large or sensitive datasets.

Sensitive data disclosure detection

Looks for exposure of PII, HR, or financial data.

Hard-coded credentials detection

Searches for embedded credentials in schemas and configurations.

Tool description injection

Analyzes tool metadata and usage descriptions to ensure safe payload handling.