mdd-stock-mcp

An MCP server for Maximum Drawdown (MDD) analytics across US stocks, Korean stocks (KOSPI/KOSDAQ), and cryptocurrencies. Plug it into Claude Desktop and ask questions like "What was SPY's worst drawdown during COVID?" or "비트코인 2022년 MDD 알려줘" — Claude will fetch the data and discuss it conversationally.

Disclaimer: This project uses yfinance, an unofficial Yahoo Finance scraper. Not affiliated with Yahoo, not for production trading systems, not for financial advice. Data may be inaccurate, delayed, or unavailable. Use at your own risk.

What it does

• Calculates Maximum Drawdown for any supported asset over a date range
• Returns OHLCV price history for plotting
• Returns a daily drawdown (underwater) series
• Compares MDD across multiple tickers side-by-side

Supported markets (via the market parameter):

Quickstart (Claude Desktop)

1. Install [uv](https://docs.astral.sh/uv/) (if you don't have it):

# macOS
brew install uv
# Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

2. Add this to `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows):

{
  "mcpServers": {
    "mdd-stock-mcp": {
      "command": "uvx",
      "args": ["mdd-stock-mcp"]
    }
  }
}

That's it. uvx fetches the package from PyPI on first run and caches it.

3. Restart Claude Desktop.

Ask: "Using mdd-stock-mcp, what was AAPL's MDD between 2020-01-01 and 2020-12-31?"

Quickstart (Claude Code CLI)

claude mcp add mdd-stock-mcp -s user -- uvx mdd-stock-mcp

Then in a new Claude Code session you'll see the mcp__mdd-stock-mcp__* tools.

Install from source (for development)

git clone https://github.com/librarywon/mdd-stock-mcp.git
cd mdd-stock-mcp
uv pip install -e ".[dev]"

Then point Claude Desktop at the local path:

{
  "mcpServers": {
    "mdd-stock-mcp": {
      "command": "uvx",
      "args": ["--from", "/absolute/path/to/mdd-stock-mcp", "mdd-stock-mcp"]
    }
  }
}

Tools

Each tool accepts:

• price: "adj_close" (default, dividend/split adjusted) or "close" (raw closing price)
• market: "us" (default), "kr", or "crypto" — see the supported markets table

Output schemas

<details> <summary><code>calculate_mdd</code> — success response</summary>

{
  "ticker": "SPY",
  "mdd_pct": -0.338915,
  "peak_date": "2020-02-19",
  "trough_date": "2020-03-23",
  "drawdown_duration_days": 33,
  "recovered": false,
  "recovery_date": null,
  "market": "us",
  "currency": "USD"
}

</details>

<details> <summary><code>calculate_mdd</code> — error response</summary>

{
  "error": true,
  "error_type": "InvalidTickerError",
  "message": "InvalidTickerError: Ticker 'XYZFAKE' not found or returned no data from Yahoo Finance."
}

</details>

<details> <summary><code>get_price_history</code> — success response</summary>

{
  "ticker": "AAPL",
  "price_basis": "adj_close",
  "count": 62,
  "data": [
    {
      "date": "2020-01-02",
      "open": 296.239990,
      "high": 300.600006,
      "low": 295.190002,
      "close": 298.829956,
      "volume": 33870100
    }
  ]
}

</details>

<details> <summary><code>get_drawdown_series</code> — success response</summary>

{
  "ticker": "SPY",
  "price_basis": "adj_close",
  "count": 125,
  "data": [
    {"date": "2020-01-02", "drawdown_pct": 0.0},
    {"date": "2020-01-03", "drawdown_pct": -0.007054},
    {"date": "2020-03-23", "drawdown_pct": -0.338915}
  ]
}

</details>

<details> <summary><code>compare_mdd</code> — success response (partial failure shown)</summary>

{
  "price_basis": "adj_close",
  "start": "2020-01-01",
  "end": "2020-06-30",
  "results": [
    {
      "ticker": "SPY",
      "mdd_pct": -0.338915,
      "peak_date": "2020-02-19",
      "trough_date": "2020-03-23",
      "drawdown_duration_days": 33,
      "recovered": false,
      "recovery_date": null,
      "error": null
    },
    {
      "ticker": "XYZFAKE",
      "mdd_pct": null,
      "peak_date": null,
      "trough_date": null,
      "drawdown_duration_days": null,
      "recovered": null,
      "recovery_date": null,
      "error": "InvalidTickerError: Ticker 'XYZFAKE' not found or returned no data from Yahoo Finance."
    }
  ]
}

</details>

How MDD is calculated

• Formula: drawdown_t = price_t / max(price_{0..t}) - 1
• MDD = min(drawdown) over the date range (most negative value)
• Peak: argmax of price up to and including the trough date
• Recovery: first date after trough where price >= peak_price
• Note: recovered=false means "did not recover within the queried window", not "never recovered historically"

All price values use dividend/split-adjusted close by default (price="adj_close"). Pass price="close" for raw unadjusted closing prices.

Example Claude conversation

User: mdd-stock-mcp으로 SPY의 2020년 코로나 폭락 분석해줘

>

Claude: [calls calculate_mdd] SPY는 2020-02-19 최고점에서 2020-03-23 최저점까지 약 -33.9%의 MDD를 기록했습니다. 회복은 2020-08-18에 완료됐어요. 그래프 그려드릴까요?

>

User: 응

>

Claude: [calls get_price_history + get_drawdown_series, draws chart in artifact]

Limitations

• US stocks, Korean stocks (KOSPI/KOSDAQ), and major cryptocurrencies — limited to what Yahoo Finance exposes. Other international exchanges (Tokyo, London, etc.) aren't wired up.
• Daily resolution — No intraday (1m, 5m, etc.) data.
• yfinance is unofficial — Yahoo Finance may rate-limit or change their API without notice. For production use, swap data.py for a paid provider like Polygon.io, Alpha Vantage, or Tiingo.
• In-memory cache only — Price data is cached for 5 minutes per process. Cache is lost on server restart.
• No real-time data — All data is end-of-day historical prices.

Development

# Install with dev dependencies
uv pip install -e ".[dev]"

# Run tests
pytest tests/

# Open MCP inspector (interactive tool tester)
fastmcp dev src/stock_mcp/server.py

Project structure:

src/stock_mcp/
  server.py   — FastMCP app + 4 tool definitions
  data.py     — yfinance wrapper + TTL cache + price-column selector
  mdd.py      — Pure MDD math functions
  models.py   — Pydantic input/output models
  errors.py   — Custom error types
tests/
  test_mdd.py — Offline math tests (no network required)

Deployment

Local (stdio) — default

The default transport is stdio, which is what Claude Desktop expects. Use the uvx config from the Quickstart section above.

Docker

# Build the image
docker build -t mdd-stock-mcp .

# Run with HTTP transport (e.g. for testing or a hosted setup)
docker run -p 8000:8000 mdd-stock-mcp --transport http

Replace 8000 with any port you prefer. The --transport http flag enables the HTTP/SSE transport mode; omit it for the default stdio mode.

Smithery / hosted

Smithery can host MCP servers so you don't need to run them locally. Once the smithery.yaml configuration is present in the repo root, you can register the server through the Smithery dashboard. See the Smithery documentation for registration instructions.

Roadmap

Planned additions (not yet scheduled — PRs welcome):

• Sharpe / Sortino / CAGR tools — additional risk/return metrics as new MCP tools, following the same patterns as the existing four.
• Optional Polygon.io / Alpha Vantage providers — drop-in replacements for the yfinance backend in data.py, configurable via environment variable.
• Shorter date ranges / intraday support — hourly or intraday resolution for ranges under ~7 days, contingent on a stable data provider.

Contributing

This is a small focused project. PRs are welcome — especially for:

• Additional data providers (Polygon, Alpha Vantage, Tiingo) in data.py
• Additional metrics (Sharpe ratio, CAGR, Sortino) as new tools
• Better error messages and edge-case handling

Please open an issue first for significant changes.

License

MIT — see LICENSE