io.github.kokogo100/ragalgo

RagAlgo: Dynamic RAG Engine for AI Reliability

"Your AI is an Analyst, NOT a Day Trader."

RagAlgo is an MCP Server that provides mathematically scored financial context (Korean Stocks/Crypto) to AI agents. We focus on "State-of-Truth" (Daily Closed Data) to prevent AI hallucinations caused by real-time market noise.

• Analyst, Not Broker: We provide "Daily Analysis Reports" (Post-Market), not real-time tick data.
• Scored Context: Instead of raw prices, we give you "Scores" (0~100) and "Zones" (Forest vs Tree).
• Global Market Specialist: Optimized for US, UK, JP, KR, and Crypto.

👉 Official Website (ragalgo.com)

---

📖 Architecture & Whitepaper

Discover why RagAlgo is the "Hippocampus" for Agentic AI, not just another RAG.

🏗️ Data Pipeline Architecture

Our production system on Railway processes global financial data 24/7:

┌─────────────────────────────────────────────────────────────────────────────┐
│                        RagAlgo Data Pipeline (Railway)                      │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  📥 COLLECT              🔍 FILTER              🏷️ TAG           📊 SCORE  │
│  ─────────────          ──────────────         ─────────        ──────────  │
│  • KR-News-Collector    • filter-worker-1      • tag-worker     • Gemini-1  │
│  • US-News-Collector    • filter-worker-2      • Meta-Hierarchy • Gemini-2  │
│  • UK-News-Collector    • filter-worker-3      •   Worker       • ...       │
│  • JP-News-Collector    • ibkr_filter_worker3  │                • Gemini-7  │
│  • research-collector   │                      │                │           │
│                         │                      │                │           │
│  ════════════════════════════════════════════════════════════════════════  │
│                                    ↓                                        │
│                      📦 SNAPSHOT (Daily 18:00 KST)                          │
│                      ──────────────────────────────                         │
│                      • KR-Snapshot  • US-Snapshot                           │
│                      • UK-Snapshot  • JP-Snapshot                           │
│                      • CRY-Snapshot • Unified-Snapshot                      │
│                                    ↓                                        │
│                          🚀 SERVE (MCP Server)                              │
│                          ─────────────────────                              │
│                          • RagAlgo-Service (SSE/stdio)                      │
│                          • ragalgo-relay-server (WebSocket)                 │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

• Vision Whitepaper (The Hook)
  • Concept: Why RagAlgo is a "Semantic Digital Twin" (SDT) using the Hippocampus analogy.
  • Value: Explains the "Self-Growing Taxonomy" and "Data Flywheel" effect.
• Technical Report (The Proof)
  • Deep Dive: Detailed anatomy of the Contextual Knowledge Network (CKN).
  • 2025 Trend: How RagAlgo serves as the memory layer for Agentic AI (e.g., PepsiCo/Salesforce Agentforce).

---

💡 Why "Daily Close"?

Users often ask: "Why isn't the chart data real-time?"

Because AI performs better with clarity. Real-time tick data is full of noise and volatility. If you feed an LLM raw live prices, it often hallucinates patterns that don't exist.

RagAlgo acts like a Professional Technical Analyst who works after the market closes:

1. Wait for the dust to settle (Market Close).
2. Analyze the day's battle (Daily Candle & Aux Indicators).
3. Deliver a "Confirmed Strategy" to your AI.

Use RagAlgo to build "Investment Advisors", not "High-Frequency Trading Bots".

---

🚀 Quick Start

Claude Desktop Configuration

Add this to your config file:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Mac: ~/Library/Application Support/Claude/claude_desktop_config.json

☁️ Cloud Mode (Recommended - No installation required)

{
  "mcpServers": {
    "ragalgo": {
      "url": "https://ragalgo-service-production.up.railway.app/sse",
      "env": {
        "RAGALGO_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

📦 Local Mode (Requires Node.js)

{
  "mcpServers": {
    "ragalgo": {
      "command": "npx",
      "args": ["-y", "ragalgo-mcp-server", "--stdio"],
      "env": {
        "RAGALGO_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

Tip: You can get a Free 1,000 Call Key instantly at RagAlgo Dashboard.

---

📚 Usage Examples (Cookbook)

We have a dedicated repository for practical examples to help you get started quickly. Please visit the RagAlgo Examples Repository.

What's Inside?

• 8 Step-by-Step Recipes: From basic data fetching to advanced AI agents.
• Skeleton Code + Prompts: Copy-paste ready resources.
• Scenarios:
  • 🐣 Basic: Get stock scores in 5 minutes.
  • 🧪 Intermediate: Verify technical signals with AI.
  • 🚀 Advanced: Build an autonomous reasoning agent (Mock Trading Audit).
  • ☕ Morning Briefing: Create a bot that emails you a daily market summary.

"Skeleton + Prompt" Approach: We provide the ingredients. You ask ChatGPT/Claude to cook!

---

🌍 Supported Markets & Roadmap

RagAlgo is expanding its CKN coverage globally. Currently, US, UK, Japan, Korea, and Crypto markets are fully supported.

Market	Asset Class	Status
🇰🇷 Korea	KOSPI / KOSDAQ	🟢 Live (Real-time Sentiment & Charts)
🇺🇸 USA	NYSE / NASDAQ	🟢 Live (Daily Scored Context)
🇯🇵 Japan	Nikkei 225	🟢 Live (Daily Scored Context)
🇬🇧 UK	LSE	🟢 Live (Daily Scored Context)
🪙 Crypto	Global (Upbit/Binance)	🟢 Live (Real-time Sentiment & Charts)

---

🛠 Tools

⚠️ CORE CONCEPT: Scored vs Raw

• get_news_scored (Default): Returns only significant news (Scores ≠ 0). Best for AI decision making.
• get_news (Raw): Returns ALL news including noise. Use this ONLY if you need raw data feed.

Tool	Description
get_news_scored	[RECOMMENDED] News WITH AI Sentiment Scores (-10 ~ +10). Filters out noise.
get_news	[Advanced] Raw News WITHOUT scores. Includes 0-score noise. Use only if you build your own scorer.
get_chart_stock	[Core] Global Stock (US/UK/JP/KR) Technical Analysis (Daily Close).
get_chart_coin	[Core] Global Crypto Technical Analysis (Daily Close).
get_snapshots	[Best] Market Overview (News + Chart + Trend) in one call.
get_financials	Corporate Financials (Quarterly/Yearly).
search_tags	Convert names (e.g., "Samsung") to RagAlgo Tags.

---

📡 Real-time WebSocket (Business Tier)

For users who really need live data (e.g., for monitoring dashboards), we offer a WebSocket stream. Note: This is strictly for monitoring, not for LLM inference context.

• Access: Business Plan subscribers (Includes 30 connections).
• Address: wss://ragalgo-relay-server-1-production.up.railway.app
• Guide: See Developer Docs for implementation details.

---

💬 Support

• Website: ragalgo.com
• Email: support@ragalgo.com