Video summary

I Built an AI Trading System With Claude + TradingView

Main summary

Key takeaways

Technology

Summary

The video shows how to build an AI-assisted pre-market trading pipeline using Claude + TradingView (via a Claude/TradingView connection), then extend and backtest with Python, and finally automate notifications to Telegram.

A key takeaway is that AI won’t make someone profitable by itself. The goal is to automate trade planning (scan, filter, backtest), while the trader’s own knowledge and oversight remain essential.

What the creator builds (end-to-end pipeline)

1) Connect TradingView charts to Claude (MCP workaround)

Goal: Allow Claude to read TradingView chart source/code so it can interpret indicators/candlestick data.

Setup details:

  • Install Claude Code and TradingView Desktop (not web TradingView).
  • MacOS requirement: PC attempts failed due to desktop file differences.
  • Uses an MCP installation copied from a GitHub repo referenced as having strong community adoption (mentioned as “2.9k stars”).
  • Includes a validation prompt: “TV health check” to confirm TradingView is connected.

Demonstration:

  • Claude identifies and marks levels such as:
    • 52-week high/low
    • daily highs/lows
    • pre-market highs/lows
  • It shows switching timeframes live and having Claude reflect those changes.

Limitations mentioned:

  • TradingView scanning appears constrained to user-selected tickers, not full-market scanning in one pass.

2) Pre-market “gap scanner” + news briefing (Scanner A)

Builds a morning workflow to find top gappers and attach a news catalyst.

Example configuration criteria:

  • Gap up > 5% vs previous day
  • Price above $3
  • Pre-market volume > 50K

Data sources and outputs:

  • Fetches ticker + gap data via Yahoo Finance
  • Gets news catalysts from Benzinga
  • Output format:
    • ticker, gap %, one-sentence catalyst
  • Saves results daily to a date-stamped JSON file
  • Claimed runtime initially: about 1–2 minutes

3) Strategy scanner filters (Scanner B) for a specific setup (Trendway Long)

Builds a TradingView-side filter that checks whether candidates meet a defined entry setup.

Example entry prerequisites (5 conditions):

  • After 10:00 a.m. market time
  • Price above yesterday’s daily high
  • Yesterday’s close above the 200 SMA
  • Price above today’s pre-market highs
  • Price above today’s high of day

Behavior:

  • If all conditions pass → scanner signals an entry list (“hit”).

Operational requirements noted:

  • TradingView Desktop must be running
  • Charting tab should be active
  • Scanner uses the tickers produced by Scanner A

4) Backtesting on TradingView with Pine Script (initial validation)

Claude generates a Pine Script strategy from the scanner criteria, then TradingView backtests it.

Demo/backtest examples:

  • Example ticker backtest on MU using a 5-minute chart
  • Reported metrics include:
    • Total P&L (example shown: ~$1,200 over 14 trades)
    • Max drawdown
    • Profit factor (example shown: ~2.48)

Caveats:

  • Validation demonstrated on limited tickers
  • Criteria described as strict, so the strategy is selective

5) Expanded backtesting using Python (more tickers / universe scaling)

Since Pine Script has scaling limits, the creator uses Claude to generate Python backtests to expand the universe.

Example scaling approach:

  • Backtest a smaller set first (e.g., “Mag 7”-style, 15-minute timeframe)
  • Then expand to a watchlist of ~32 tickers

Python backtest example described:

  • Lookback: last 30 days
  • Example results:
    • ~54% win rate
    • Total P&L shown for a simulated $10,000 account
  • Some tickers may show no signals (e.g., Meta, Robinhood)

6) Automation: scheduled scanning + Telegram notifications

The pipeline is automated so scanning results trigger without manual prompting.

Scheduling logic

  • Scanner A (pre-market)
    • Start 8:30 a.m.
    • Run every 30 minutes
    • Continue until 2:00 p.m.
    • Example: 12 runs/day
  • Scanner B (strategy filter)
    • Starts after 10:00 a.m.
    • Runs about 5 minutes after pre-market updates
    • Continues until 3:00 p.m. (implied by scheduling)

Scanner B runs are tied to the tickers produced by Scanner A.

Telegram integration

  • Uses Telegram BotFather to create a bot
  • Workflow requirement:
    • The user provides a message format so bot outputs are consistent (Scanner A results + Scanner B “hits”)
  • Demonstrated phone outputs:
    • Morning: “pre-market gappers preview” (ticker, gap %, news catalyst)
    • After open: “trend joint long watch list” (strategy hits)
  • Screenshot claim: the bot triggers when new scanner runs complete.

Bonus: TradingView “AI chart co-pilot” (beta) via TradingView Remix

The creator introduces TradingView Remix (Chrome extension) and TradingView’s public beta AI chart co-pilot.

Claimed benefit:

  • May simplify certain tasks and reduce limitations observed in the Claude + MCP approach.

Demonstration:

  • Backtest on a watchlist for the last 90 days on a 15-minute timeframe

Observed limitations:

  • The comparison suggests behavior may differ from earlier workflows (one demo indicated it “not working” as expected).

Additional test described:

  • Ask AI to generate a top-10 pre-market gap scan by 8:30 a.m. using criteria like:
    • price > $3
    • pre-market volume threshold
    • gap %
    • catalyst

Limitations specifically called out:

  • Can’t schedule automatically; must be run on demand
  • In the demo, it likely can’t send to Telegram directly (automation/integration not shown as supported)

Key product/feature comparison & takeaways

  • Claude + TradingView MCP connection

    • Strong for:
      • generating prompts
      • parsing chart-related data
      • producing Pine Script strategies
    • Limited for:
      • market scanning breadth
      • scalable backtesting across large universes in TradingView/Pine

  • Python extension

    • Addresses scale better:
      • more tickers (e.g., ~32 watchlist)
      • broader universe backtests

  • Automation

    • Scanner results can be delivered to Telegram reliably via bot setup

  • TradingView Remix AI chart co-pilot (beta)

    • Useful for AI-assisted analysis/backtesting attempts
    • Constraints noted:
      • scheduling may not be supported
      • Telegram integration may not work out-of-the-box

Main speakers / sources (as indicated)

  • Primary speaker/source: “Shay” (creator/trader; repeatedly referenced as a full-time trader with 10+ years experience)
  • Referenced tools/sources:
    • Claude / Claude Code (claude.ai + “Claude Code”)
    • TradingView Desktop
    • GitHub MCP workaround repo
    • Yahoo Finance (gap % and pre-market values)
    • Benzinga (news catalyst sourcing)
    • TradingView Remix / AI chart co-pilot (beta) (via Chrome extension)
    • Telegram BotFather
    • Interactive Brokers API (mentioned as a next step toward a fully automated trading system)

Original video