Tradr

Async trading toolbox + MCP server for AURA. Includes market scans, portfolio/risk analytics, trade execution, and lightweight memory (watchlist + notes).

Disclaimer

This software can place live orders if configured with your broker API key.

It is for research/experimentation. Running it against real money is entirely at your own risk. This is not financial advice or a managed service.

Quick start

Install deps:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements-mcp.txt

Set environment:

cp .env.example .env

Run MCP server (stdio):

python mcp_server.py

Run HTTP MCP server:

python http_mcp_server.py

Environment variables

See .env.example for the full list. Required:

• ALPACA_KEY
• ALPACA_SECRET
• GOOGLE_API_KEY
• FINNHUB_KEY

Optional:

• SEC_API_KEY
• EARNINGSFEED_API_KEY

MCP tools (high level)

Research / scanning:

• get_market_news
• get_latest_sec_filings
• assess_market_regime
• scan_top_movers
• scan_unusual_options_flow
• scan_insider_activity (EarningsFeed)
• scan_13f_changes (EarningsFeed)

Technicals:

• analyze_technicals
• analyze_volume_trend

Portfolio / risk:

• get_portfolio_state
• analyze_sector_exposure
• calculate_risk_metrics
• calculate_portfolio_beta

Execution:

• smart_order_entry (bracket orders)
• modify_trade_parameters (stop updates)
• close_position
• buy_option

Watchlist + notes:

• add_to_watchlist
• get_watchlist (returns fresh technicals per symbol)
• remove_from_watchlist
• save_notes / read_notes

Storage locations

• Notes: notes.md (sliding window of last 10 entries).
• MCP memory (watchlist, trades, etc.): /tmp/agent_memory.json.
• CLI memory: agent_memory.json in repo root.

Tests

Tool suite:

python run_tests.py --type tools

1. Claude Desktop

Add this to your claude_desktop_config.json (usually ~/Library/Application Support/Claude/claude_desktop_config.json on Mac):

{
  "mcpServers": {
    "aura-trading": {
      "command": "/absolute/path/to/venv/bin/python",
      "args": ["/absolute/path/to/tradr/mcp_server.py"],
      "env": {
        "ALPACA_KEY": "your_key",
        "ALPACA_SECRET": "your_secret",
        "GOOGLE_API_KEY": "your_key",
        "FINNHUB_KEY": "your_key"
      }
    }
  }
}

Note: We recommend using absolute paths to your venv python executable. Also, ensure the Python environment has mcp installed.

2. Gemini Code Assist / Editor

Add this to your project's .gemini/settings.json (or global settings):

{
  "mcpServers": {
    "tradr": {
      "command": "/absolute/path/to/venv/bin/python",
      "args": ["/absolute/path/to/tradr/mcp_server.py"],
      "cwd": "/absolute/path/to/tradr",
      "timeout": 30000,
      "trust": true
    }
  }
} 

Notes

I personally suggest using this with Gemini 3 Pro. In my personal experience Gemini 3 outperformed Opus 4.5 - this is strictly my personal experience and should not be taken as financial advice or a recommendation.

I have not yet tested GPT 5.2 or Grok with this toolset.

How to Vibe trade

Wire up the MCP server to your AI agent of choice, set the environment variables, and start vibing.

Example prompt:

"Hmm google earnings today, stock down a lil I know:

• anthropic trained opus 4.5 on tpus
• cost reduction likely due to tpus
• inference on tpus
• msft cloud disappointment
• openai previously also eyed tpus for their models

its basically the if transformers are a thing, google is the sauce, and i am extremely bullish on them, obviously doesnt mean the market agrees - but earnings are after hours today, opinion?"

Agent will look up options flows, insider activity, and 13f changes, company news, and technicals to make a informed suggestion or YOLO your money its an LLM you never know - beware risks THERE ARE NO GUARDRAILS IN PLACE.

LICENSE

MIT