Short answer
Give the model two tools, one for live quotes and one for price history, and let it decide when to call them. The Anthropic Python SDK’s tool runner handles the loop. Data comes from Yahoo Finance Turbo with a RapidAPI key, so it works from servers where yfinance gets rate-limited.
The whole agent
import os
from datetime import datetime, timezone
import anthropic
import requests
from anthropic import beta_tool
HOST = "yahoo-finance-turbo.p.rapidapi.com"
HEADERS = {"X-RapidAPI-Key": os.environ["RAPIDAPI_KEY"], "X-RapidAPI-Host": HOST}
def turbo(path: str, **params) -> dict:
r = requests.get(f"https://{HOST}{path}", params={"region": "US", **params}, headers=HEADERS, timeout=30)
r.raise_for_status()
return r.json()
@beta_tool
def get_quotes(symbols: str) -> str:
"""Latest price, day change, volume and market cap for up to 20 stocks.
Args:
symbols: Comma-separated Yahoo Finance symbols, e.g. "NVDA,AMD".
"""
rows = turbo("/market/v2/get-quotes", symbols=symbols)["quoteResponse"]["result"]
keep = ("symbol", "shortName", "currency", "regularMarketPrice", "regularMarketChangePercent",
"regularMarketVolume", "marketCap", "fiftyTwoWeekRange", "regularMarketTime")
return str([{k: q.get(k) for k in keep} for q in rows])
@beta_tool
def get_price_history(symbol: str, range: str = "6mo", interval: str = "1d") -> str:
"""Daily or weekly closing prices for one stock, as "date,close" lines.
Args:
symbol: One Yahoo Finance symbol, e.g. "NVDA".
range: How far back: 1mo, 6mo, 1y, 5y or max.
interval: Bar size: 1d, 1wk or 1mo.
"""
r = turbo("/stock/v3/get-chart", symbol=symbol, range=range, interval=interval)["chart"]["result"][0]
closes = r["indicators"]["quote"][0]["close"]
lines = [f"{datetime.fromtimestamp(t, timezone.utc):%Y-%m-%d},{c:.2f}" for t, c in zip(r["timestamp"], closes) if c is not None]
return "\n".join(lines)
client = anthropic.Anthropic() # reads ANTHROPIC_API_KEY
runner = client.beta.messages.tool_runner(
model="claude-opus-5-5",
max_tokens=16000,
betas=["server-side-fallback-2026-07-01"],
fallbacks="default",
system=(
"You are a stock research assistant. Get every number from the tools and say which "
"date it is from. Show your arithmetic. Do not recommend buying or selling."
),
tools=[get_quotes, get_price_history],
messages=[{"role": "user", "content": "Compare NVDA and AMD over the last 6 months: total return, "
"largest drop from a peak, and today's move."}],
)
final = runner.until_done()
print("".join(b.text for b in final.content if b.type == "text"))pip install anthropic requests
export ANTHROPIC_API_KEY=... # console.anthropic.com
export RAPIDAPI_KEY=... # Yahoo Finance Turbo on RapidAPI
python stock_agent.pyHow it works
@beta_toolturns each function’s signature and docstring into a tool definition, so the docstring is what the model reads to decide when to call it.- The tool runner sends the question, runs whatever tools Claude asks for, sends back the results and repeats until Claude answers.
- Tools return small text, not raw API JSON. Six months of daily closes as
date,closelines is about 3 KB, which keeps the model’s context and your token bill small. - The system prompt makes the agent cite the date of each number and show its arithmetic, and keeps it from giving trade advice.
fallbacks="default"lets the API retry on another Claude model if the first one declines a request.
Give it more tools
Each extra tool is one more short function over a Turbo endpoint. Good next ones:
| Tool | Endpoint | Returns |
|---|---|---|
| Fundamentals | /stock/get-fundamentals | Income statement, balance sheet and cash flow |
| Valuation | /stock/v4/get-statistics | P/E, margins, beta and other key statistics |
| Earnings | /stock/get-earnings | Earnings history and estimates |
| Analysts | /stock/get-recommendation-trend | Recommendation trend over time |
| Options | /stock/v3/get-options | Option chains with strikes and expiries |
| Screener | /screeners/get-symbols-by-predefined | Day gainers, most active and other preset screens |
The Yahoo Finance API page lists all 37 endpoints. Pass only the fields the agent needs. Smaller tool results make answers cheaper and more accurate.
Questions it handles well
- “How did the S&P 500 (^GSPC) do this year compared with QQQ?”
- “Which of AAPL, MSFT and GOOGL is closest to its 52-week high?”
- “Plot NVDA weekly closes for 5 years and describe the three biggest drawdowns.” (Add a plotting tool or use code execution.)
Not financial advice. Yahoo Finance Turbo is a third-party API, not affiliated with Yahoo.
Yahoo Finance Turbo
37 endpoints, a stated quota, free plan of 500 requests a month.