yfinance-mcp

Live market data for your AI. A Model Context Protocol server that gives Claude (or any MCP client) real-time quotes, OHLCV history, full options chains, news, and multi-ticker comparison. Free, no API key.

<div align="center"> <img src="assets/demo.svg" alt="Claude answering a market question through yfinance-mcp" width="860"> </div>

Why

LLMs answer market questions with training-data guesses. Wire this server in once and they stop guessing:

• "What's Apple trading at?" → live quote instead of a stale hallucination
• "Pull AAPL options expiring next month" → real strikes with IV and open interest
• "NVDA vs AMD vs INTC this year?" → actual returns, volatility, drawdowns, computed from data
• "Any recent news on PLTR?" → headlines with links it can cite

Tools

Tool	What it does	Example prompt
get_quote	Last price, day range, volume, market cap, 52w range, change %	"what's TSLA doing today?"
get_history	OHLCV bars (any period × interval) + summary stats	"SPY daily closes for 2024"
get_options_chain	Calls & puts: strike, last, bid/ask, IV, OI, volume, ITM	"AAPL chain for next month"
get_news	Recent headlines with links	"any news on PLTR?"
compare_tickers	Return, annualized vol, max drawdown side by side; names the winner	"MSFT vs GOOGL, 3 months"

Responses are JSON, truncated to LLM-friendly sizes (history capped at 250 bars, options at 60 rows/side), with summary statistics so the model reasons over aggregates before raw bars.

Setup

git clone https://github.com/saksham10arora-dotcom/yfinance-mcp
cd yfinance-mcp
python3 -m venv .venv && .venv/bin/pip install -r requirements.txt

Claude Desktop / any MCP client

Add to claude_desktop_config.json (File → Settings → Developer → Edit Config):

{
  "mcpServers": {
    "yfinance": {
      "command": "/absolute/path/to/yfinance-mcp/.venv/bin/python",
      "args": ["/absolute/path/to/yfinance-mcp/server.py"]
    }
  }
}

Restart Claude Desktop. The tools appear under the 🔨 icon. Ask about any ticker.

Verify without an MCP client

./.venv/bin/python -c "import server; print(server.get_quote('AAPL'))"

CI runs the same stdio handshake GitHub does on every push, so main is always protocol-correct.

FAQ

Is data real-time? Yahoo Finance data: near-real-time for US equities, delayed for some international exchanges. Good for research and chat, not for order execution.

Does it cost anything? No. No API key, no rate-limit account, nothing to configure.

Why not just paste yfinance code into chat? You can. This removes the copy-paste loop and lets the model decide which tool to call per question, including multi-step ones ("compare these five tickers, then pull the chain on the most volatile").

Windows? Works. Point the config at .venv\Scripts\python.exe and use double backslashes in JSON.

Troubleshooting

• Tool icon missing after restart: paths in the config must be absolute; relative paths silently fail.
• no data for <ticker>: delisted or misspelled symbol; check Yahoo Finance in a browser.
• Rate limiting after heavy use: back off a few minutes; Yahoo throttles bursts.

Alternatives compared

	yfinance-mcp	Paid market MCPs
API key	none	required
Cost	free	$10-100+/mo
Options chains with IV	yes	varies
Real-time depth/L2	no	sometimes
Setup time	2 min	account + billing

License

MIT. Not investment advice; the AI reading your portfolio is not a fiduciary.