MCP Server Configuration

MCP Server Configuration for Trading Platform

Experiment Overview

Current Configuration

The project uses .mcp.json in the repository root for MCP server configuration:

{
  "mcpServers": {
    "alpaca": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--with", "pytz", "alpaca-mcp-server", "serve"],
      "env": {
        "ALPACA_API_KEY": "your_key",
        "ALPACA_SECRET_KEY": "your_secret",
        "ALPACA_PAPER_TRADE": "True"
      }
    },
    "sqlite-trading": {
      "type": "stdio",
      "command": "uvx",
      "args": ["mcp-sqlite", "C:/Users/smith/Alpaca_trading/data/trading.db"]
    },
    "sqlite-market-data": {
      "type": "stdio",
      "command": "uvx",
      "args": ["mcp-sqlite", "C:/Users/smith/Alpaca_trading/data_cache/market_data.db"]
    }
  }
}

Configured Servers

1. Alpaca MCP Server (alpaca-mcp-server)

Purpose: Trade stocks, ETFs, crypto, and options via natural language through Claude Code.

Features:

• Portfolio management and account information
• Real-time market data and historical data
• Order placement, cancellation, and modification
• Watchlist management

Known Issue: The alpaca-mcp-server package has a missing pytz dependency when run via uvx. Fix by adding --with pytz to args.

Source: https://github.com/alpacahq/alpaca-mcp-server

2. SQLite MCP Servers (mcp-sqlite)

Purpose: Query trading databases directly from Claude Code.

Databases:

• sqlite-trading: Trading records (data/trading.db)
• sqlite-market-data: Market data cache (data_cache/market_data.db)

Usage: Natural language queries like "Show me all trades from last week" or "What's the schema of the predictions table?"

Source: https://github.com/jparkerweb/mcp-sqlite

Failed Attempts

Recommended Additional MCP Servers

For Trading Platform Development

Example: Adding Yahoo Finance for Sector Data

"yahoo-finance": {
  "type": "stdio",
  "command": "uvx",
  "args": ["yahoo-finance-mcp"]
}

WARNING: Yahoo Finance MCP is ONLY for sector/industry lookups. NEVER use for OHLCV price/volume data - use Alpaca API exclusively for that.

Troubleshooting

Missing Dependency Errors

ModuleNotFoundError: No module named 'xyz'

Fix: Add "--with", "xyz" to the args array before the package name.

Server Won't Start

1. Test manually: uvx --with pytz alpaca-mcp-server --help
2. Check API keys are set correctly
3. Restart Claude Code after changing .mcp.json

API Key Security

• Never commit actual API keys to .mcp.json
• Use environment variables or separate key files
• Paper trading keys shown in examples are for demonstration only

Key Insights

1. uvx isolates environments - Each MCP server runs in its own isolated Python environment
2. --with flag for deps - Add missing transitive dependencies with --with package_name
3. Restart required - Claude Code must restart to pick up .mcp.json changes
4. Multiple SQLite servers - Can configure separate servers for different databases
5. Data source policy still applies - MCP servers don't change the rule: Alpaca API for OHLCV, yfinance only for sector lookups

Files Modified

References

• Alpaca MCP Server: https://github.com/alpacahq/alpaca-mcp-server
• Alpaca MCP Docs: https://docs.alpaca.markets/docs/alpaca-mcp-server
• mcp-sqlite: https://github.com/jparkerweb/mcp-sqlite
• MCP Protocol: https://modelcontextprotocol.io/
• Skill: data-source-priority - Data source policy (Alpaca API mandatory)