TradingView MCP (Headless Market Data)

Market data via the bundled tradingview MCP server — TradingView's public scanner API plus Yahoo Finance, spoken over MCP. No TradingView desktop app or account. Most tools need no API key. The server ships with the full plugin (mcp.json for portable/Codex packages, .mcp.json for legacy clients) and starts when enabled in a host that supports local stdio MCP processes. Its 39 tools appear under the plugin's tradingview MCP namespace (load them through the client's deferred-tool mechanism when necessary). Installing only the skill does not install the MCP server; use the full finance-data-providers plugin or configure the server separately.

Read-only. Nothing here places trades or mutates any account.

When to use this vs the other data skills

Rule of thumb: this skill first for anything price/TA/scan shaped — it needs zero setup and won't force a CDP relaunch of the user's TradingView app. Drop to tradingview-reader only for greeks or account-bound data (watchlists, alerts, charts).

Step 1: Check requirements

• uv installed (brew install uv) — the server runs via uvx and self-installs on first launch (give the first call ~30s).
• Python 3.10–3.13. The pinned upstream currently excludes Python 3.14.
• Optional: MARKETAUX_API_TOKEN enables market_sentiment, financial_news, and the news legs of combined analysis.

If the MCP tools are missing, ask the user to restart the agent after enabling the plugin. Do not silently replace market data with invented values.

Step 2: Select the smallest useful tool set

Prefer one focused call for a simple question. For a research brief, combine only the independent views that matter: price/snapshot, TA, volume, and optionally news. Avoid calling multiple overlapping TA tools merely to produce more output.

Quotes & snapshots

Technical analysis (single symbol)

Exchange-wide scans

Futures (夜盘 / overnight)

Options (US equities — no greeks)

Backtesting

News & sentiment (needs MARKETAUX_API_TOKEN)

market_sentiment(symbol, category) · financial_news(symbol, category, limit)

Regional extras

egx_* (Egyptian Exchange suite, incl. egx_smart_money_scanner) — rarely relevant; also BIST, HKEX, SSE, SZSE, TWSE supported via the exchange param on scan tools.

Step 3: Normalize inputs

1. exchange defaults to KUCOIN (crypto!) on most scan/TA tools — always pass NASDAQ / NYSE explicitly for US equities.
2. Three symbologies coexist: scanner tools take bare symbol + exchange param; stock_prices takes EXCHANGE:SYMBOL; Yahoo-backed tools (yahoo_price, backtests, options) take Yahoo tickers (BTC-USD, ^VIX, THYAO.IS).
3. Timeframes: 5m, 15m, 1h, 4h, 1D, 1W, 1M. Intraday scans on stock exchanges can be sparse — prefer 1D for equities.
4. Tool names are canonical as listed. Use coin_analysis and multi_timeframe_analysis, not README-style aliases such as get_technical_analysis.

Step 4: Validate and interpret results

1. Check timestamps, exchange, symbol, session, and currency before drawing conclusions.
2. Treat a returned error envelope as an error, not as an empty result. Use its retryability and symbol suggestions; retry at most once when marked retryable.
3. Filter wide scanner results to roughly 10 rows and the columns relevant to the question.
4. Treat Yahoo options IV as suitable for chain shape and OI/volume positioning, not precise IV rank or skew. Cross-check IV-sensitive work with tradingview-reader.
5. Backtests are historical simulations. Report the period, interval, costs, sample size, benchmark, and walk-forward result when available; do not present them as forecasts.

Step 5: Respond to the user

Lead with the answer, then show a compact evidence table. Include an as-of timestamp and data source, name any missing or stale fields, and separate observed facts from interpretation. For trading-shaped requests, keep the response analytical and read-only: discuss risks and scenarios without placing or offering to place a trade.