tasty-agent: A TastyTrade MCP Server

Overview

A Model Context Protocol server for interacting with TastyTrade brokerage accounts. This server enables Large Language Models to monitor portfolios, analyze positions, and execute trades through the TastyTrade platform.

Please note that tasty-agent is currently in early development. The functionality and available tools are subject to change and expansion as development continues.

Prerequisites

• Python 3.12
• uv package manager
• A TastyTrade account

Installation

Install uv if you haven't already:

curl -LsSf https://astral.sh/uv/install.sh | sh

We will use uvx to directly run tasty-agent:

uvx tasty-agent

Authentication

The server requires TastyTrade credentials. For security, these are set up via command line and stored in your system's keyring (Keychain on macOS, Windows Credential Manager on Windows, or similar secure storage on other platforms):

tasty-agent setup

Tools

Portfolio Management

1. plot_nlv_history

   • Plots account net liquidating value history over time
   • Input:
     • time_back (string): Time period to plot ('1d', '1m', '3m', '6m', '1y', 'all')
   • Returns: Base64-encoded PNG image of the generated plot

2. get_account_balances

   • Get current account balances
   • Returns: Formatted string with cash balance, buying power, net liquidating value, and maintenance excess
   • Example response: "Cash: $5,000.00, Buying Power: $10,000.00, NLV: $15,000.00"

3. get_open_positions

   • Get all currently open positions
   • Returns: Formatted table showing Symbol, Position Type, Quantity, and Current Value

4. get_transaction_history

   • Get transaction history
   • Input:
     • start_date (string, optional): Start date in YYYY-MM-DD format
   • Returns: Formatted table showing Transaction Date, Transaction Type, Description, and Value

Trade Management

1. schedule_trade

   • Schedule a trade for execution
   • Inputs:
     • symbol (string): Stock or option symbol
     • quantity (integer): Number of shares/contracts
     • action (string): "Buy to Open" or "Sell to Close"
     • execution_type (string): "immediate", "once", or "daily"
     • run_time (string, optional): Time to execute in HH:MM format (24-hour), defaults to market open (09:30)
     • dry_run (boolean): Simulate without executing
   • Returns: Task ID and confirmation message
   • Notes:
     • Immediate trades during market hours execute right away
     • Immediate trades during market closure are scheduled for next market open
     • Daily trades repeat every market day at the specified time
     • One-time trades execute once at the specified time

2. list_scheduled_trades

   • List all scheduled trades
   • Returns: Formatted table showing Task ID, Time, Type, and Description

3. remove_scheduled_trade

   • Remove a scheduled trade
   • Input:
     • task_id (string): ID of task to remove
   • Returns: Confirmation message

Market Analysis

1. get_metrics

   • Get market metrics for specified symbols
   • Input:
     • symbols (string[]): List of stock symbols
   • Returns: Formatted table showing IV Rank, IV Percentile, Beta, Liquidity Rating, and Next Earnings Date

2. get_prices

   • Get current bid and ask prices
   • Input:
     • symbol (string): Stock or option symbol
   • Returns: Current bid and ask prices

Usage with Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "tastytrade": {
      "command": "uvx",
      "args": ["tasty-agent"]
    }
  }
}

Important: The tasty-agent server runs as a background process managed by Claude Desktop. Scheduled trades will only execute while Claude Desktop is running. When Claude Desktop is closed, the server gracefully saves all scheduled tasks and resumes them when Claude Desktop is reopened. Tasks are stored securely in ~/.tasty_agent/scheduled_tasks.json.

Debugging

You can use the MCP inspector to debug the server:

npx @modelcontextprotocol/inspector uvx tasty-agent

For logs, check:

• macOS: ~/Library/Logs/Claude/mcp*.log
• Windows: %APPDATA%\Claude\logs\mcp*.log

Development

For local development testing:

1. Use the MCP inspector (see Debugging)
2. Test using Claude Desktop with this configuration:

{
  "mcpServers": {
    "tastytrade": {
      "command": "uv",
      "args": [
        "--directory",
        "path/to/tasty-agent",
        "run",
        "tasty-agent"
      ]
    }
  }
}

Security Notice

This server handles sensitive financial information and can execute trades. Always:

• Use secure credential storage
• Review queued orders before execution
• Use dry-run mode for testing
• Monitor scheduled tasks regularly
• Keep your Claude Desktop installation up to date

License

This MCP server is licensed under the MIT License. See the LICENSE file for details.