Forensics Connection MCP Server
A Model Context Protocol (MCP) server that analyzes forensic evidence to identify people and their connections, then generates Python code for network visualization.
Features
- LLM-Powered Relationship Analysis: Uses OpenAI's GPT-4 to intelligently analyze relationships between people in evidence text
- Network Visualization: Generates comprehensive Python code using NetworkX and Matplotlib for interactive network graphs
- Evidence Parsing: Automatically extracts and categorizes different types of evidence (police reports, witness statements, communications, etc.)
- Connection Strength Scoring: Assigns relationship strength scores (1-10) based on evidence quality
- Cluster Analysis: Identifies groups and central figures in the network
- Flexible Filtering: Configurable minimum connection strength thresholds
Installation
- Clone the repository:
git clone <repository-url>
cd mcp-forensics
- Install dependencies:
npm install
- Set up environment variables:
cp .env.example .env
# Edit .env and add your OpenAI API key
- Build the project:
npm run build
Configuration
Create a .env file with your OpenAI API key:
OPENAI_API_KEY=your_openai_api_key_here
Usage
As an MCP Server
The server can be used with any MCP-compatible client:
npm start
Available Tools
analyze_connections
Analyzes evidence text to identify people and their connections.
Parameters:
evidence(required): The evidence text to analyzeoptions(optional):includeVisualization(boolean, default: true): Whether to include Python visualization codeminimumConnectionStrength(number, default: 3): Minimum connection strength to include (1-10)groupByOrganization(boolean, default: true): Whether to group people by organization
Returns:
- List of people identified with their organizations and roles
- Connections between people with strength scores and evidence
- Central figures in the network
- Clusters/groups of related people
- Python code for network visualization
Example Usage
// Example evidence input
const evidence = `
Police Report #001: Incident occurred at 123 Main St on January 15, 2024 at 2:30 PM.
Witness John Smith reported suspicious activity. Vehicle license plate ABC123 was seen leaving the scene.
Email Chain - Subject: Concerns about OpenAI:
From: elon.musk@x.com To: satya.nadella@microsoft.com, mark.zuckerberg@meta.com
"I'm increasingly concerned about Sam's strategic direction..."
Meeting Notes - Private Tech Leaders Dinner:
Attendees: Musk, Nadella, Zuckerberg. Topic: OpenAI concerns.
`;
// The server will return detailed analysis and Python visualization code
LLM-Enhanced Relationship Analysis
The system uses OpenAI's GPT-4 to analyze relationships with sophisticated context understanding:
- Direct Communications: Emails, messages, calls (strength: 9-10)
- Face-to-Face Meetings: Shared events, professional collaboration (strength: 7-8)
- Group Communications: Shared concerns, indirect interactions (strength: 5-6)
- Professional Associations: Mentioned together, weak connections (strength: 3-4)
- Coincidental Mentions: Very weak connections (strength: 1-2)
Python Visualization Output
The generated Python code includes:
- Network Graph: Interactive visualization with NetworkX and Matplotlib
- Organization Grouping: Color-coded nodes by organization
- Connection Types: Different edge styles for different relationship types
- Centrality Analysis: Node sizes based on degree centrality
- Statistical Report: Network statistics and key findings
- Customizable Layout: Spring layout for optimal node positioning
Required Python Dependencies
The generated code requires:
pip install networkx matplotlib numpy
Development
Scripts
npm run build: Compile TypeScript to JavaScriptnpm run dev: Build with watch modenpm start: Run the compiled servernpm run inspector: Run with MCP inspector for debugging
Project Structure
src/
├── index.ts # Main server entry point
├── types.ts # TypeScript type definitions
├── evidenceParser.ts # Evidence parsing and LLM analysis
├── pythonGenerator.ts # Python code generation
└── ...
Connection Types
The system identifies several types of connections:
- communication: Direct communications (emails, messages)
- meeting: In-person meetings and events
- witness: Witness/observer relationships
- location: Geographic/location-based connections
- organization: Same organization affiliations
- other: General associations and mentions
Error Handling
- Fallback to pattern-based analysis if LLM calls fail
- Rate limiting protection for OpenAI API calls
- Comprehensive error reporting and logging
- Graceful degradation for missing data
License
MIT License