Skip to main content

​
Prerequisites

  • Cursor, Claude Code, or any MCP-compatible client

​
1. Connect your MCP client

Add the following to your MCP client configuration. For Cursor — open ~/.cursor/mcp.json: 
{
  "mcpServers": {
    "quantspace": {
      "url": "https://mcp-fabric-server.blackmushroom-7d2b97e6.eastus2.azurecontainerapps.io/mcp"
    }
  }
}
Restart your client. All nine tools will appear automatically. No login or token required. See the Clients section for client-specific instructions.

​
2. Run your first pipeline

Once connected, ask your AI assistant:
“Run the full ML trading pipeline for AAPL, MSFT, GOOGL from 2023-01-01 to 2024-01-01, test until 2025-01-01”
The agent will call the core pipeline tools in order: 
run_data_extraction → run_feature_worker → run_ml_job → run_po_job → run_trading_job
Optional post-processing tools after run_po_job: 
run_plot_job | run_st_job | run_risk_job
Or run each step manually. Start with data extraction: 
{
  "config": {
    "backtest_params": {
      "Tickers": "custom",
      "Data_source": "yahoo",
      "Learning_start": "2023-01-01",
      "Learning_end": "2024-01-01",
      "Testing_end": "2025-01-01"
    },
    "custom": ["AAPL", "MSFT", "GOOGL"]
  }
}
The tool returns an output_url — pass it to run_feature_worker as input_url.

​
3. Resume from an existing result

Every stage writes its output to blob storage. To skip a stage, pass an existing blob URL directly to the next tool as input_url. Browse the data container in Storage Explorer to find URLs of previous outputs. Blob names follow the pattern <prefix><timestamp>.json, for example: 
data_extractor_20260307_124937.json
feature_engine_20260307_125103.json