marc-shade/fraud-detection-mcp

このページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。原文を見る

6 stars 3 forks 品質 71 トレンド 14MIT

Fraud detection MCP server: Isolation Forest + Autoencoder ensemble, behavioral biometrics, agent-to-agent transaction protection, defense compliance (NITTF/SIEM/SEAD). 24 tools, 831 tests.

概要

Part of the Agentic System - a 24/7 autonomous AI framework with persistent memory. - : Anomaly detection for real-time transaction screening - : PyTorch-based deep learning anomaly detection using reconstruction error scoring. Ensemble with Isolation Forest (configurable 60/40 weighting). - : Keystroke dynamics (Isolation Forest), mouse movement patterns (One-Class SVM), touch screen patterns (LOF) - : NetworkX-based graph centrality metrics (degree, clustering coefficient, betweenness, closeness) for fraud ring detection - : Comprehensive feature engineering with cyclical encoding, z-scores, and velocity features - : Automatic detection of AI agent vs human traffic (Stripe ACP, Visa TAP, Mastercard Agent Pay, Google AP2, PayPal, Coinbase, OpenAI, Anthropic, x402) - : API key format validation, JWT token expiry checks, JSON-backed agent registry - : Per-agent Isolation Forest baselines (8 features, max 1000 observations/agent) - : Spending limits, merchant whitelists/blocklists,...

README

以下は原文 README の抜粋です。原文を見る。

Advanced Fraud Detection MCP

Fraud detection and anomaly analysis for financial security — with AI agent-to-agent transaction protection.

Part of the Agentic System - a 24/7 autonomous AI framework with persistent memory.

Overview

An open-source Model Context Protocol (MCP) server for fraud detection using machine learning and behavioral analysis. This system combines behavioral biometrics, anomaly detection, network graph analysis, and AI agent behavioral fingerprinting for fraud prevention.

Key Features

Core Detection (Active by Default)

Isolation Forest: Anomaly detection for real-time transaction screening
Autoencoder Ensemble: PyTorch-based deep learning anomaly detection using reconstruction error scoring. Ensemble with Isolation Forest (configurable 60/40 weighting).
Behavioral Biometrics: Keystroke dynamics (Isolation Forest), mouse movement patterns (One-Class SVM), touch screen patterns (LOF)
Network Graph Analysis: NetworkX-based graph centrality metrics (degree, clustering coefficient, betweenness, closeness) for fraud ring detection
46-Feature Pipeline: Comprehensive feature engineering with cyclical encoding, z-scores, and velocity features

Agent-to-Agent Transaction Protection

Traffic Classification: Automatic detection of AI agent vs human traffic (Stripe ACP, Visa TAP, Mastercard Agent Pay, Google AP2, PayPal, Coinbase, OpenAI, Anthropic, x402)
Agent Identity Verification: API key format validation, JWT token expiry checks, JSON-backed agent registry
Agent Behavioral Fingerprinting: Per-agent Isolation Forest baselines (8 features, max 1000 observations/agent)
Mandate Compliance: Spending limits, merchant whitelists/blocklists, time windows, geographic restrictions
Collusion Detection: Directed graph analysis for circular flows, temporal clustering, and volume anomalies
Agent Reputation Scoring: Longitudinal trust from history, consistency, and collusion safety (weighted 40/25/25/10)

Defense Insider Threat Compliance

28 Behavioral Indicators from the NITTF Insider Threat Guide (EO 13587)
SIEM Integration: CEF (ArcSight), LEEF (QRadar), Syslog RFC 5424 with MITRE ATT&CK enrichment
Cleared Personnel Monitoring: SEAD 4/6 continuous evaluation, 13 adjudicative guidelines

GitHub で README 全文を見る

インストール

This server does not publish a one-line install command.

Open the repository installation guide MCP 初心者？mcpskill.org のガイドから始めましょう →

設定

{
  "mcpServers": {
    "fraud-detection-mcp": {
      "command": "/path/to/fraud-detection-mcp/venv/bin/python",
      "args": ["/path/to/fraud-detection-mcp/server.py"],
      "env": {
        "FRAUD_DETECT_MODEL_PATH": "/path/to/fraud-detection-mcp/models",
        "FRAUD_DETECT_LOG_LEVEL": "INFO"
      }
    }
  }
}