About this project

This repository presents a Crypto MCP Server and Hedgehog Agentic Trading Dashboard, an autonomous hedge fund AI orchestrator developed by Corax CoLAB. The system is built on the Model Context Protocol (MCP) infrastructure, designed to serve as a localized, AI-powered backbone for local LLMs such as Claude Desktop to ingest real-time market data, analyze technicals and sentiment, and autonomously execute trades. The core differentiator is its multi-agent architectural approach. Rather than being a simple API wrapper, the system operates as a multi-agent playground with zero mock-ups and 100% real data across all visualizers, driven by live websockets and REST data. The architecture natively multiplexes JSON-RPC commands and supports easy extensibility through new MCP modules. The Autonomous Orchestrator Mode implements a continuous Observe-Analyze-Act (OODA) loop running 24/7. The Observe phase queries multiple MCPs for technical indicators, news sentiment, and on-chain data. The Analyze phase evaluates signals using LLMs (Gemini, Claude, or OpenAI) to produce structured BUY, SELL, or HOLD decisions. The Act phase executes trades via the local CCXT MCP. The system also includes Agentic Backtesting for safely testing AI prompts and models against historical OHLCV data without risking real funds, and a Proof of Brain trading diary that generates comprehensive markdown reports detailing the reasoning behind every action. A Telegram Command Center provides real-time alerts and remote control capabilities, with commands for viewing active AI providers and last decisions, receiving Proof of Brain reports, and triggering manual OODA cycles. The dashboard features multiple real-time data visualizations powered by React Three Fiber and dynamic emissive materials: a Dark Pool Sonar for large volume whale trades, a Flash-Crash Prediction Matrix tracking liquidity drains via CCXT orderbooks, a Galaxy View mapping the top 50 cryptocurrencies in 3D, an AI Sentiment Word-Cloud Sphere extracting trending keywords from real-time news, a Gas and Network Congestion Hologram visualizing Ethereum network congestion, and a Time-Machine Backtest Arena for OHLCV visualization and strategy simulation. The Multi-MCP Ecosystem includes over a dozen MCP servers covering exchange trading and market data (CCXT), on-chain data (ETH balances, ERC20, gas prices), technical analysis (RSI, MACD, SMA, Bollinger Bands, Monte Carlo simulations), portfolio management, market statistics (CoinGecko), LLM operations, notifications and Telegram broadcasting, and integrations with Freqtrade, OctoBot, Hummingbot, and Superalgos platforms. The system uses a centralized .env configuration file for managing Python MCP servers and the Node.js backend, with security features including password-protected backend execution, fail-closed pair restrictions, maximum trade amount limits, and crypto.timingSafeEqual for password verification. Orders are logged to a local SQLite database. Installation is supported through an automated install.sh script or manual setup, with systemd service management for the backend. The frontend is built with Vite and React, while the backend uses Node.js. The project requires Python 3.11+ and Node.js 20+. Security best practices emphasized include using testnet keys for testing, keeping API keys out of the repository, restricting network access to local LAN, and implementing authenticated reverse proxies.