oarjones/mcp-ibkr-trader

mcp-ibkr-trader

Autonomous trading system connecting a Master Control Program (MCP) with Interactive Brokers Gateway.

Quick Start

1. Install dependencies:

   pip install -r requirements.txt
   

2. Run the server:

   uvicorn mcp_server.main:app --reload
   

3. Run tests:

     pytest -q
   

4. Example usage:

   curl -s -X POST http://localhost:8000/tool/market_data.get_bars \
     -H 'Content-Type: application/json' \
     -d '{"symbol":"EUR.USD","asset_type":"FX","tf":"1m","start":"2025-08-01T07:00:00Z","end":"2025-08-01T07:30:00Z"}'
   

Configuration

The system can be configured via config.example.yaml. Key settings include:

• ibkr.market_data.hist_defaults.outside_rth: A boolean flag to control whether historical data requests include data outside Regular Trading Hours (RTH).
  • If true, useRTH is set to 0 (data outside RTH is included).
  • If false, useRTH is set to 1 (only RTH data is included).

IBKR Adapter Details

The ibkr_adapter module includes several refinements for robust interaction with the Interactive Brokers TWS API:

• Real-time Subscription Management:

  • Teardown on Disconnect: Active real-time market data and real-time bars subscriptions are automatically cancelled when the TWSClient disconnects, preventing orphaned subscriptions and resource leaks.
  • Resubscription on Reconnect: Upon successful reconnection to the TWS Gateway, the adapter attempts to re-establish any real-time market data or real-time bars subscriptions that were active prior to the disconnection. This ensures continuity of data streams.

• Historical Data Concurrency:

  • get_historical_data calls are limited to a maximum of 2 concurrent requests using a threading semaphore. This helps to prevent pacing violations with the IBKR API. If more than 2 requests are made simultaneously, subsequent requests will wait or raise a TimeoutError if the semaphore cannot be acquired within the specified timeout.

• get_bars DataFrame dtypes:

  • The get_bars method in ibkr_adapter/adapter.py ensures consistent data types for the returned Pandas DataFrame:
    • open, high, low, close: float64
    • volume: Int64 (nullable integer)
    • ts (timestamp): datetime64[ns] (timezone-naive) This guarantees data quality for subsequent analytical operations.