tradesdontlie/tradingview-mcp
This MCP server gives Claude Code local chart, Pine Script, alert, and screenshot controls through TradingView Desktop, but depends on undocumented internals and may conflict with platform terms.
The project connects Claude Code to TradingView Desktop through an MCP server, local **CDP port 9222**, and **84 MCP tools**. It supports chart navigation, indicator reads, Pine Script iteration, drawings, alerts, replay, screenshots, layouts, and JSONL monitoring.
It is a useful reference for making a stateful Electron application legible to an agent: expose compact structured reads before screenshots, keep commands available through both MCP and CLI, and add explicit health checks and tool-selection guidance.
The project connects Claude Code to TradingView Desktop through an MCP server, local **CDP port 9222**, and **84 MCP tools**. It supports chart navigation, indicator reads, Pine Script iteration, drawings, alerts, replay, screenshots, layouts, and JSONL monitoring. It is a useful reference for making a stateful Electron application legible to an agent: expose compact structured reads before screenshots, keep commands available through both MCP and CLI, and add explicit health checks and tool-selection guidance. It requires a valid TradingView subscription and deliberate debug-port activation. The integration uses undocumented internal interfaces that may break, cannot execute trades, and its programmatic data access may conflict with TradingView terms or market-data licenses.
TradingView MCP makes a stateful desktop application agent-readable through compact structured operations, CLI parity, health checks, and selective screenshots. It specializes the browser-control pattern into a deep application harness, but also exposes a sharper adoption boundary: useful automation depends on a debug port, undocumented internals, subscription access, and acceptable terms and data-license use, while trade execution remains out of scope.