tejas-040303/tradingview-mcp

Purpose and process topology

This repository exposes two independent local MCP servers for AI-assisted trading workflows: a TradingView Desktop controller/reader over Chrome DevTools Protocol (CDP), and a read-only MetaTrader 5 data service backed by Python. They can run separately, communicate only through localhost, and are also usable through a tv CLI.

• src/server.js

• src/mt5-server.js

• src/cli/index.js

TradingView MCP boundary

The TradingView server is a Node.js stdio MCP process. Tool registration/adaptation lives in src/tools/, while operational implementations live in src/core/. A shared CDP connection layer targets the user-launched TradingView Electron instance on port 9222; TradingView itself remains the stateful chart/UI authority.

• src/connection.js

• src/core/index.js

• src/tools/chart.js

TradingView workflow domains

Core modules organize chart state/data extraction, chart and pane navigation, Pine editor automation, drawings/alerts, replay, screenshots, UI automation, watchlists, tabs, and polling streams. Typical control flow is MCP tool/CLI command → tool adapter → core operation → CDP evaluation/UI interaction → compact structured response.

• src/core/chart.js

• src/core/pine.js

• src/core/stream.js

CLI and command routing

The tv CLI is a parallel front door rather than a separate implementation: command modules parse user input and route into the same capabilities, preserving pipe-friendly JSON behavior. The router/index are the primary CLI composition points.

• src/cli/index.js

• src/cli/router.js

• src/cli/commands/pine.js

MT5 read-only bridge

Because MT5’s official binding is Python and Windows-specific, the Node MT5 MCP server proxies to a separate local HTTP bridge. The bridge owns terminal calls, normalization, analytics, calendar ingestion, and strict read-only route behavior; no order-operation path exists. MT5 times are normalized to explicitly labeled broker-server and UTC values, with null/refusal rather than guessed offsets.

• src/mt5-server.js

• mt5-bridge/bridge.py

• mt5-bridge/mt5_client.py

Broker analytics and strategy simulation

The Python subsystem converts raw deals into paired trades, derives excursion/analytics insights, and provides deterministic strategy setup detection, backtests, parameter sweeps, reconciliation against actual trades, and reconstructed paper state. Simulation output is intentionally shaped like real trades so downstream analytics/dashboard use a common path.

• mt5-bridge/normalize.py

• mt5-bridge/analytics.py

• mt5-bridge/strategy.py

Dashboard and runtime orchestration

The bridge serves a same-origin dashboard and API. A React/Vite analytics UI consumes precomputed API statistics rather than calculating metrics client-side; plain HTML/JS dashboard fallbacks coexist under the bridge. The Node launcher probes services, starts only missing local processes, and conditionally builds dashboard assets.

• dashboard-app/src/App.jsx

• dashboard-app/src/lib/api.js

• scripts/start.js

External dependencies and infrastructure

Material runtimes are Node.js for MCP, CLI, CDP control, and launcher; Python 3.9+ plus MetaTrader5 for the broker bridge; and a build-time React/Vite/Tailwind dashboard. External application boundaries are TradingView Desktop via local CDP :9222, MT5 terminal through the Python package, and an MQL5 calendar exporter that writes JSON for the bridge to consume.

• mt5-bridge/calendar_export.mq5

• mt5-bridge/calendar_export.mq5

• dashboard-app/vite.config.js