MCPWatch

Coordinates 11 specialized vulnerability detection scanners through the MCPScanner orchestrator class using a pipeline pattern that manages repository cloning, parallel scanner execution, result aggregation, and cleanup operations. Each scanner extends an AbstractScanner base class providing common utilities for credential sanitization, file system operations, and result formatting, enabling modular vulnerability detection across MCP server implementations.

Implements CredentialScanner that detects hardcoded API keys, tokens, and insecure credential storage patterns in MCP server code using pattern matching against known credential formats (AWS keys, OpenAI tokens, private keys, etc.). The scanner includes built-in credential sanitization utilities in the AbstractScanner base class to mask sensitive data in reports, preventing accidental exposure of discovered secrets.

Executes all 11 vulnerability scanners in parallel using asynchronous operations, aggregating results from each scanner into a unified report. The orchestrator manages concurrent execution to balance performance with resource utilization, collecting vulnerability objects from each scanner and merging them by category and severity for comprehensive reporting.

Provides AbstractScanner base class with shared utilities including credential sanitization, file system operations, result formatting, and error handling. All specialized scanners extend this base class to inherit common functionality, reducing code duplication and ensuring consistent vulnerability reporting across all scanner implementations. Utilities include regex-based pattern matching, file reading, and credential masking.

Implements ToolPoisoningScanner that detects hidden malicious code, suspicious function implementations, and tool poisoning attacks in MCP server tool definitions. The scanner analyzes function signatures, implementation patterns, and data flow to identify code that may exfiltrate data, execute arbitrary commands, or bypass security controls through the MCP tool interface.

Implements scanners that detect parameter injection vulnerabilities, improper input validation, and MCP protocol violations in server implementations. The detection engine analyzes how MCP servers handle tool parameters, resource requests, and protocol messages to identify injection attack vectors, missing validation, and deviations from the MCP specification that could enable exploitation.

Integrates vulnerability detection patterns derived from authoritative security research sources including the VulnerableMCP database, HiddenLayer research on parameter injection attacks, and Trail of Bits credential security analysis. The system maps research findings to specialized scanner implementations, enabling detection of known MCP vulnerability categories with patterns informed by real-world attack research and security best practices.

Implements configurable severity filtering (critical, high, medium, low) and vulnerability category filtering that allows users to focus scan results on relevant threats. The reporting system aggregates vulnerabilities by category and severity, providing both detailed findings and summary statistics. Users can filter results before or after scanning to customize output based on risk tolerance and compliance requirements.

Generates vulnerability reports in multiple output formats including structured JSON for programmatic consumption and human-readable console output with formatting. The CLI interface routes scan results to the appropriate formatter based on user selection, enabling integration with downstream tools, dashboards, and reporting systems while maintaining readability for direct terminal use.

Handles secure cloning of GitHub repositories to temporary directories for local analysis, with automatic cleanup after scanning completes. The system manages file system operations including directory creation, repository cloning via Git, file reading for vulnerability analysis, and cleanup of temporary artifacts. Error handling ensures cleanup occurs even if scanning fails, preventing disk space leaks.

Implements an extensible scanner system where each vulnerability detector extends the AbstractScanner base class, providing common utilities and enforcing a consistent interface. New scanners can be added by extending the base class and registering with the MCPScanner orchestrator. The architecture enables community contributions and custom vulnerability detection without modifying core orchestration logic.

Implements command-line interface using Commander.js library to parse user arguments, validate inputs, and route scan requests to the MCPScanner orchestrator. The CLI supports multiple options including repository URL, output format, severity filtering, and category filtering. Argument parsing includes validation and helpful error messages for invalid inputs.