Alpaca Trading Bot (MCP Server)

An MCP server exposing Alpaca paper trading as tools for Claude (or any MCP client): check quotes, manage a watchlist, check market hours, and place simulated trades. No real money is ever at risk — paper=True is hardcoded in trading_mcp_server.py.

Setup

python -m venv .venv
.venv/Scripts/activate   # or source .venv/bin/activate on macOS/Linux
pip install -r requirements.txt
cp .env.example .env     # then fill in your Alpaca paper trading keys

Get paper trading API keys from the Alpaca dashboard (alpaca.markets) — sign up, then generate a key under the Paper Trading section (not Live).

Running

python trading_mcp_server.py

This runs as a stdio MCP server. To register it with Claude Code:

claude mcp add alpaca-trading -- python /full/path/to/trading_mcp_server.py

Tools

• get_account — cash, buying power, portfolio value

• get_quotes(symbols?) — latest bid/ask for given tickers, or the whole watchlist if omitted

• get_watchlist / add_to_watchlist(symbol) / remove_from_watchlist(symbol) — manage watchlist.json

• is_market_open — checks Alpaca's trading calendar (accounts for weekends + market holidays)

• list_positions — current open positions

• place_order(symbol, qty, side) — market order, paper trading only

• list_orders — recent order history

Watchlist

watchlist.json seeds the top 25 QQQM (Nasdaq-100) holdings by weight, verified via stockanalysis.com. Edit the file directly, or use the add_to_watchlist / remove_from_watchlist tools.

Not yet built

• Scheduled runs (e.g. every 30 minutes during market hours): the MCP server itself only responds to on-demand tool calls — it does not self-schedule. Options discussed but not decided: a local script + Task Scheduler/cron, or an Azure Function with a timer trigger (Azure Functions has GA support for MCP tool triggers as of Build 2026 — see functions-bindings-mcp-tool-trigger docs). is_market_open is the building block for whichever scheduler wraps it.

• Autonomous vs. approval-gated trading: not yet decided whether the AI should place orders on its own or require confirmation each time.