Yahoohoo

Yahoohoo is an MCP server for market-data analysis. It uses Yahoo Finance for equities, ETFs, and options, CoinGecko for crypto data, and sandboxed Python for numerical computation. No API keys are required for the default setup.

It works with any MCP host, including Claude, Codex, ChatGPT, Cursor, and Windsurf. When the host supports MCP sampling, Yahoohoo can break larger requests into smaller analysis steps and synthesize the results. When it does not, Yahoohoo falls back to deterministic heuristics.

That recursive analysis looks like this in practice:

compare QQQ vs IWM vs TLT performance and risk
├─ analyze QQQ
├─ analyze IWM
├─ analyze TLT
└─ synthesize cross-symbol comparisons and conclusions

Install

npm install -g yahoohoo
yahoohoo --transport stdio

Requires Node 22+ and Python 3.

Host Setup

Claude Code

claude mcp add --transport stdio yahoohoo -- npx yahoohoo --transport stdio

Codex

codex mcp add yahoohoo -- npx yahoohoo --transport stdio

Claude Desktop — add this to claude_desktop_config.json:

{
  "mcpServers": {
    "yahoohoo": {
      "command": "npx",
      "args": ["yahoohoo", "--transport", "stdio"]
    }
  }
}

ChatGPT and other remote hosts — run Yahoohoo in HTTP mode behind HTTPS. See docs/CHATGPT.md.

Available tools

Market data (Yahoo Finance)

Tool	What it does
yahoo_search	Find instruments by name or symbol
yahoo_quote	Current quotes for up to 20 symbols
yahoo_historical	OHLCV bars, or summary=true for computed stats
yahoo_quote_summary	Fundamentals and company profile
yahoo_options	Option chains with multi-expiry sweep
yahoo_screener	day_gainers, day_losers, most_actives, and other predefined screens
yahoo_trending	Trending symbols by region
yahoo_insights	Technical levels, outlook, and analyst recommendations

Crypto (CoinGecko)

Tool	What it does
crypto_trending	Trending coins and categories
crypto_global	Total market cap, BTC/ETH dominance, and volume
crypto_markets	Top coins by market cap with price changes
crypto_coin	Detailed coin info, including supply, ATH/ATL, and categories
crypto_search	Search coins and get CoinGecko IDs

Yahoo Finance also supports crypto tickers like BTC-USD and ETH-USD for quotes and historical data. The CoinGecko tools fill in the gaps with trending data, market-cap rankings, market-wide metrics, and richer coin metadata.

Analysis tools

Tool	What it does
yahoohoo_analyze	Run a sandboxed analysis and return a summary with a runId
yahoohoo_peek	Explore saved analysis results by dot-path
yahoohoo_drill	Run follow-up analysis while inheriting the previous run's parameters
yahoohoo_options_dashboard	Render an ASCII options dashboard with GEX, key levels, and dealer-positioning proxies

The analysis engine classifies intent (for example comparison, risk, options, fundamentals, or macro), selects the matching Python program, fetches the required data, and runs the computation in a sandboxed subprocess. For more involved questions it can split the work into child analyses and store the detailed output server-side, while the host receives a compact summary and a runId for deeper inspection.

Optional: llm_query() inside the sandbox

Sandbox Python code can call llm_query("prompt") to make a local LLM sub-call during computation (RLM pattern). Yahoohoo auto-detects which supported CLI is installed:

CLI	Command	Install
Claude	claude -p "prompt"	claude.ai/code
Codex	codex exec "prompt"	github.com/openai/codex
Gemini	gemini -p "prompt"	github.com/google-gemini/gemini-cli
Copilot	copilot -p -s "prompt"	github.com/github/copilot-cli

Detection order is claude, then codex, then gemini, then copilot. Override it with:

YAHOOHOO_LLM_CLI=gemini    # force a specific CLI
YAHOOHOO_LLM_CLI=auto      # auto-detect (default)
YAHOOHOO_LLM_QUERY=false   # disable entirely

This only works when the CLI is available on the same machine as Yahoohoo. It does not apply when Yahoohoo is running remotely for a host like ChatGPT or Claude Desktop.

Example

yahoohoo_analyze({
  "query": "compare QQQ vs IWM vs TLT performance and risk",
  "symbols": ["QQQ", "IWM", "TLT"]
})

Analysis objective: compare QQQ vs IWM vs TLT performance and risk
Scope: QQQ, IWM, TLT
Window: 2025-09-08 → 2026-03-07 (1wk)

Comparison snapshot:
- best total return: IWM (5.27%)
- lowest volatility: TLT (17.23%)
- shallowest drawdown: TLT (-4.85%)
- total return: IWM (5.27%) > QQQ (2.23%) > TLT (-1.66%)
- risk-adjusted return: IWM (0.16) > QQQ (0.07) > TLT (-0.10)

Pairwise:
- QQQ vs IWM: corr=0.65, rel.ret=-3.03%
- QQQ vs TLT: corr=-0.03, rel.ret=3.89%

[run: abc123] [confidence: 0.85] [intent: comparison]

Documentation

• Architecture — three-layer design, data flow, and transport strategy
• Configuration — environment variables, HTTP mode, and deployment
• Security — sandbox threat model and limitations
• ChatGPT Setup — running in HTTP mode for remote hosts

Caveats

• Yahoo Finance is unofficial, so data availability and response shapes can change.
• Options dealer-position outputs such as GEX and gamma flip are estimates derived from open interest and Black-Scholes gamma, not direct dealer inventory.
• CoinGecko's free tier is limited to roughly 10-30 calls per minute without a key and 30 calls per minute with a demo key (COINGECKO_API_KEY).
• The Python sandbox blocks dangerous imports and applies resource limits, but it is not VM-level isolation. See Security for details.

Development

npm run check         # syntax-check source files
npm test              # 213 deterministic tests, no network
npm run test:live     # includes live Yahoo Finance integration
npm run smoke:sandbox # smoke test the Python sandbox

License

MIT