PLUTUS Open Source - Breaking the Barrier in Algorithmic Trading

Zero-Setup Market Data Analytics with Python API, CLI, and LLM Integration

PLUTUS is a data analytics framework for Vietnamese stock market with three ways to access 21GB of historical data (2021-2022): Python API, command-line tools, and natural language queries through LLM integration.

What is PLUTUS?

PLUTUS provides zero-setup access to Vietnamese market data without database installation:

• 📊 Rich Dataset: 21GB tick & daily data from HSX, HNX, UPCOM (2000-2022)
• 🚀 Zero Setup: Query CSV files directly using DuckDB (no database required)
• ⚡ High Performance: Optional Parquet optimization for 10-100x faster queries
• 🔧 Triple Interface: Python API + CLI + LLM integration (MCP)
• 🤖 AI-Powered: Query data using natural language through Claude, Gemini, or other MCP clients
• ✅ Production Ready: 205+ tests, comprehensive documentation

Quick Start

Installation

git clone https://github.com/algotradevn/plutus.git
cd plutus
pip install -e .

Configuration

Set your dataset path (choose one method):

Option 1: Environment Variable (Recommended)

export HERMES_DATA_ROOT=/path/to/hermes-offline-market-data-pre-2023

Option 2: Config File

cp config.cfg.template config.cfg
# Edit config.cfg and set PLUTUS_DATA_ROOT

First Query

Python API:

from plutus.datahub import query_historical

# Get 5-minute OHLC bars
ohlc = query_historical(
    ticker_symbol='FPT',
    begin='2021-01-15',
    end='2021-01-16',
    type='ohlc',
    interval='5m'
)

for bar in ohlc:
    print(f"{bar['bar_time']}: O={bar['open']} H={bar['high']} "
          f"L={bar['low']} C={bar['close']}")

CLI:

python -m plutus.datahub \
  --ticker FPT \
  --begin 2021-01-15 \
  --end 2021-01-16 \
  --type ohlc \
  --interval 5m \
  --output fpt.csv

LLM (Natural Language):

> Get me FPT's 5-minute OHLC bars for January 15, 2021

Features

1. DataHub Library (Python API)

Programmatic access to market data with flexible querying:

Tick Data Queries:

from plutus.datahub import query_historical

# Get tick-level data with field selection
ticks = query_historical(
    ticker_symbol='HPG',
    begin='2021-01-15 09:00:00',
    end='2021-01-15 10:00:00',
    type='tick',
    fields=['matched_price', 'matched_volume', 'bid_price_1', 'ask_price_1']
)

for tick in ticks:
    print(f"{tick['datetime']}: {tick['matched_price']} @ {tick['matched_volume']}")

OHLC Aggregation:

# Generate candlestick bars from tick data
ohlc = query_historical(
    ticker_symbol='VIC',
    begin='2021-01-15',
    end='2021-01-16',
    type='ohlc',
    interval='15m',  # 1m, 5m, 15m, 30m, 1h, 4h, 1d
    include_volume=True
)

Features:

• 40+ data fields (matched price/volume, bid/ask, foreign flows, open interest)
• 7 OHLC intervals (1m, 5m, 15m, 30m, 1h, 4h, 1d)
• Date/datetime range filtering
• Lazy iteration for memory efficiency
• DataFrame conversion via to_dataframe()

📖 Python API Documentation

2. DataHub CLI

Command-line interface for data export and analysis:

# Export tick data to CSV
python -m plutus.datahub \
  --ticker FPT \
  --begin "2021-01-15 09:00" \
  --end "2021-01-15 10:00" \
  --type tick \
  --fields matched_price,matched_volume \
  --output fpt_ticks.csv

# Generate OHLC bars in JSON format
python -m plutus.datahub \
  --ticker HPG \
  --begin 2021-01-15 \
  --end 2021-01-16 \
  --type ohlc \
  --interval 1m \
  --format json \
  --output hpg_1m.json

# Get query statistics before execution
python -m plutus.datahub \
  --ticker VIC \
  --begin 2021-01-01 \
  --end 2021-12-31 \
  --stats

Output Formats: CSV, JSON, table (terminal)

📖 CLI Usage Guide

3. MCP Server (LLM Integration)

Access market data through natural language using Claude Desktop, Gemini CLI, or other MCP-compatible LLMs.

What is MCP?

Model Context Protocol (MCP) enables LLMs to access external data sources through a standardized interface. Instead of writing code, you query data using natural language.

Quick Setup

1. Start MCP Server:

python -m plutus.mcp

2. Configure Your Client:

{
  "mcpServers": {
    "plutus-datahub": {
      "command": "python",
      "args": ["-m", "plutus.mcp"],
      "env": {
        "HERMES_DATA_ROOT": "/absolute/path/to/dataset"
      }
    }
  }
}

claude mcp add --transport stdio plutus-datahub python -- -m plutus.mcp

npm install -g @google/gemini-cli@latest
gemini auth login

gemini mcp add plutus-datahub python -m plutus.mcp \
  -e HERMES_DATA_ROOT=/absolute/path/to/dataset \
  --description "Vietnamese market data access"

gemini
> @plutus-datahub Get FPT's daily OHLC for January 15, 2021

3. Query with Natural Language:

Try these queries in your MCP client:

• Basic Data: "Get FPT's daily OHLC data for January 2021"
• Intraday Analysis: "Show me VIC's 5-minute OHLC bars on Jan 15, 2021 with volume"
• Tick Data: "Get HPG's matched price and volume from 9am to 10am on Jan 15"
• Comparison: "Compare FPT and VIC performance for Q1 2021"
• Technical Analysis: "Calculate RSI and MACD for HPG in January 2021"
• Anomaly Detection: "Find unusual volume spikes for FPT in 2021"

MCP Features

• 4 Tools: query_tick_data, query_ohlc_data, get_available_fields, get_query_statistics
• 4 Resources: Dataset metadata, ticker list, field descriptions, OHLC intervals
• 5 Prompts: Daily trends, volume analysis, ticker comparison, anomaly detection, technical indicators

Supported Clients

• ✅ Claude Desktop (macOS, Windows)
• ✅ Claude Code (VS Code extension)
• ✅ Gemini CLI (Terminal, all platforms)
• ✅ Custom MCP Clients (Python/TypeScript SDK)

📖 MCP Documentation:

• Quick Start Guide - 5-minute setup
• Client Setup - Detailed configuration for all clients
• Tools Reference - Complete API documentation
• Usage Examples - Real-world query examples

Dataset

Plutus requires the hermes-offline-market-data-pre-2023 dataset (~21GB):

• Coverage: 2021-2022 (2 years)
• Exchanges: HSX, HNX, UPCOM
• Data Types: Tick-level intraday + daily aggregations
• Format: CSV files (optionally convert to Parquet for 10-100x faster queries)

📧 Contact ALGOTRADE for dataset access

Performance Optimization

Out of the box, Plutus queries CSV files directly (zero setup). For production use:

# Convert to Parquet (10-100x faster, 60% smaller)
python -m plutus.datahub.cli_optimize optimize --data-root /path/to/dataset

Benefits:

• 10-100x faster queries
• 60% smaller storage footprint
• Metadata caching for instant field lookups

📖 Performance Guide

Requirements

• Python: 3.12 or higher
• Dataset: hermes-offline-market-data-pre-2023 (21GB)
• Dependencies: Automatically installed via pip
  • DuckDB (query engine)
  • PyArrow (Parquet support)
  • FastMCP (MCP server)
  • Others (see pyproject.toml)

Project Status

• Version: 1.0.0 (October 2025)
• Tests: 205/205 passing ✅
• Production Ready: DataHub + MCP Server

Current Features:

• ✅ DataHub (Python API + CLI)
• ✅ MCP Server (Claude Desktop, Gemini CLI, custom clients)
• ✅ Performance optimization (Parquet, metadata cache)
• 🚧 Trading algorithms (Framework in development)

Architecture

Plutus follows the ALGOTRADE 9-step algorithmic trading process:

1. Define trading hypothesis
2. Data collection ← DataHub provides this layer ✅
3. Data exploration
4. Signal detection
5. Portfolio management
6. Risk management
7. Backtesting
8. Optimization
9. Live trading

The DataHub module (production-ready) handles step 2 with three interfaces:

• Python API for programmatic access
• CLI for data export and batch processing
• MCP Server for LLM integration

Other modules are under development.

Documentation

DataHub

• CLI Usage Guide - Command-line examples and workflows
• Performance Optimization - Parquet conversion and tuning
• Python Examples - Ready-to-run Python scripts

MCP Server

• Quick Start - 5-minute setup for Claude/Gemini
• Client Setup - Detailed configuration guide
• Tools Reference - Complete API documentation
• Usage Examples - Query patterns and workflows
• Setup Scripts - Server setup and integration

Troubleshooting

Dataset Not Found

Error: Dataset not found at: /path/to/dataset

Solution: Set HERMES_DATA_ROOT environment variable or edit config.cfg

Import Errors

ModuleNotFoundError: No module named 'plutus'

Solution: Install in development mode: pip install -e .

Slow Queries

Solution: Convert data to Parquet format (see Performance Guide)

MCP Connection Issues

Solution: See MCP Quick Start for client-specific troubleshooting

Contributing

This is a research project. For questions or collaboration:

• GitHub Issues: https://github.com/algotradevn/plutus/issues
• Email: andan@algotrade.vn

License

MIT License - See LICENSE file for details.

Author

Dan (andan@algotrade.vn) ALGOTRADE - Algorithmic Trading Education & Research

Acknowledgments

Built on the ALGOTRADE 9-step methodology for systematic algorithmic trading development.