---

2. Demo

GIF placeholder: docs/media/demo.gif — Claude Desktop receiving a screenshot_chart() of an AAPL daily, then calling extract_indicators() and giving a plain-language read. (Record a 12–20s loop: ask "what's on my screen?" → Claude shows the chart back + the RSI/EMA/MACD read.)

3. Install

# 1) install the server (pipx keeps it isolated)
pipx install claude-tradingview-mcp        # or: uv tool install claude-tradingview-mcp

# 2) register it with Claude
claude mcp add tradingview claude-tradingview-mcp

Then grant macOS Accessibility + Screen Recording to the app running the server (Terminal / Claude Desktop) — see docs/install.md.

4. Quickstart

Drop the snippet from examples/claude-desktop-config.json into your Claude Desktop config, chart any symbol in TradingView, and ask:

"What's on my screen?"

Claude calls get_symbol() → screenshot_chart() → extract_indicators() and tells you what it sees.

5. Why

Claude can already analyze prose. Now it can analyze your charts — the exact ones on your screen, with your indicators and your layout. The chart stays on your machine: the screenshot is captured from your window, handed to Claude, and the temp file is deleted immediately. Nothing is uploaded by this tool, and nothing it exposes can place a trade.

6. The tools (all read-only)

Tool	What it returns
get_symbol()	the charted symbol + timeframe (from the window title)
screenshot_chart()	a base64 PNG of only the TradingView window
extract_indicators()	RSI, EMA20/50/200, MACD, ATR, Bollinger, VWAP, Supertrend + a verdict
get_watchlist()	symbols TradingView currently exposes
read_alerts()	read-only view of visible alert configs
resource tradingview://current	current symbol + open windows snapshot

Tools this server will never have

place_order · modify_alert · paper_trade · withdraw · buy · sell · cancel_order · close_position · transfer — registration of any of these raises at boot. See safety.py and tests/test_safety.py.

7. Demo video

Placeholder: a 60-second screencast (link here). Show the boot-time safety abort when a write tool is added — that's the moment that sells the project.

8. FAQ

Will it ever execute a trade? No. There is no code path that can. Tools are registered through a hardcoded allowlist in safety.py; any tool whose name even contains a trade/write verb raises SafetyViolation before the server starts. The test suite asserts the registry is exactly the five read-only tools.

Does my chart get uploaded anywhere? The screenshot is captured locally and deleted from disk right after it's encoded. This tool uploads nothing. (Your MCP client — Claude — receives the image to analyze; that's the point.)

Does it work on Windows/Linux? Not yet — it relies on macOS AppleScript + screencapture. PRs welcome for a Windows/Linux read path, as long as they stay read-only.

Do I need a TradingView API key / paid plan? No. There is no TradingView API here — it reads the desktop app's window. Indicator math uses free public OHLC from yfinance.

9. Comparison

	claude-tradingview-mcp	ccxt MCP servers	tradingview-screener
Reads your actual on-screen chart	✅	❌	❌
Can place orders	❌ never	✅ (that's the point)	❌
Needs API keys / exchange creds	❌	✅	❌
Screenshot of your live window	✅	❌	❌
Indicator bundle	✅ (local, MIT)	varies	scanner-based
Safety model	read-only by construction	you hold the keys	read-only

10. Contributing

PRs welcome — with one hard rule: we will not merge any PR that adds a write tool (orders, alert mutation, money movement, "just paper trading", anything). The CI safety test will fail it, and so will we. See docs/threat-model.md.

11. License & sponsors

MIT © 2026 Aiden Hecker. See LICENSE. Indicator formulas are pure pandas (no AGPL dependency). If this saves you time, sponsorship keeps it maintained.

12. Requirements

macOS · TradingView desktop app · Python 3.10+ · Accessibility + Screen Recording permission for the host app. How it works: docs/how-it-works.md.

---