AI vs Trading · Claude Series

Claude × TradingView
the AI that codes your Pine Script indicators(and what it will never do)

The Captain connected Claude, Anthropic's AI, to TradingView. The result: Pine Script indicators coded in a few sentences, in plain English, without writing a single line of code yourself — plus, new: the full 13-minute tutorial to wire up the same setup at home. But let's be clear from the start: the AI assists, it doesn't trade for you. This page shows you both sides, videos included.

The tutorials + the lives + the shorts

Claude × TradingView: the official MCP

Zero lines of code written by hand

At a glance

The essentials

01
AI trading or trading AI — same idea, four different things: rule-based bots (grid, DCA, Expert Advisors), quantitative models, generative AI used as a copilot (ChatGPT, Claude, Gemini…), and regulated robo-advisors. Four families, four risk profiles: do not lump them together.
02
No AI can predict how markets will move with certainty. AI saves you time on research and writing code; reading the market and making the call stay with the trader.
03
What we actually tested, with videos to prove it: Claude connected to TradingView via an MCP server — Pine Script indicators without coding, alerts, watchlists, scanning hundreds of assets. The same bridge to MetaTrader 5 is documented in 6 steps, in a written guide on this page. And TradingView now has its own official MCP connector: 4 steps, nothing to install.
04
It costs money and time: a paid Claude subscription, a paid TradingView subscription, tokens that run out fast, and a setup that can be a pain.
05
An "AI trading" tool that guarantees a high return with no risk: walk away — there is a strong chance it is a scam. Before you put any money in, check that the platform is authorized by the relevant regulator.

New · the official connector

Official TradingView MCP: it's finally here!

TradingView has launched its own MCP server, in public beta since September 16, 2026: no more terminal, no more TradingView Desktop, no more debug port. Paste a URL into Claude or ChatGPT, authorize access to your account, and your AI works with your watchlists, your alerts, the screener and the economic calendar. The essentials unfold right here, in one click; the complete step-by-step — 35 tools, beta limits, security, troubleshooting — is in our TradingView MCP guide.

Full guide · free access

Connect TradingView to your AI, in 4 steps

TradingView's official connector, nothing to install: a URL to paste into Claude, ChatGPT or Codex, then you authorize access to your account. It never places an order, but it can create or modify your watchlists and your alerts. Essential plan or higher, trials excluded.

Before you start

You'll need a TradingView Essential plan or higher, not a trial: trials don't include MCP access. On the AI side: Claude (on the web, the desktop app or mobile — even on the free plan, which is limited to one custom connector), ChatGPT or Codex. Nothing to install: the server runs on TradingView's end, no need for TradingView Desktop.

1
Add the connector
- ChatGPT / Codex (desktop app): Settings → Plugins → Add → MCPs → Add → Add MCP Server.
- Claude: Settings → Connectors → Add custom connector (depending on your version: Customize → Connectors → “+”).
In Claude, a connector added once works everywhere: on the web, in the desktop app and on mobile.
2
Copy-paste these details
```
Name: TradingView Official MCP Server
URL: https://mcp.tradingview.com/mcp
```
If you're asked for a connection type, choose Streamable HTTP. You can name it whatever you like: plain "TradingView", as in the tutorial, works too.
3
Connect your account
Save, click Authenticate (in Claude: Add, then Connect), sign in to TradingView and authorize access. The connection happens on TradingView's official page (OAuth): you never give your password to your AI, and there's no API key to copy. Tip from the tutorial: sign in to TradingView in your browser first, and authorization then takes just one click.
4
Run your first test
```
Use TradingView to show me my watchlists.
```
What you should see: the list of your TradingView watchlists. If your AI asks for permission to use the tool, accept — it's a simple read, nothing gets changed.

Once connected — prompts to try

1.“Use TradingView to show me my watchlists.”
2.“Using TradingView, calculate the Choppiness Index (14) on H4 for the assets in my Crypto watchlist and keep the ones above 60.”
3.“In my TradingView, create a watchlist called Chop 4h with these assets.”
4.“Using TradingView’s screener, list the cryptos whose volume today is more than twice their 10-day average.”
5.“Get me the 15-minute BTCUSDT candles from yesterday afternoon and tell me what the price did after my 2:30pm entry.”
6.“Create a TradingView alert if BTC falls back below $100,000: that’s my stop level.”
7.“Using TradingView, which high-impact macro announcements are scheduled this week in the United States?”

Tip: start your requests with “Using TradingView” or “Use TradingView” — otherwise your AI may go and fetch its data somewhere else. Also say whether you mean spot or futures: when that wasn't specified, the tutorial's scan defaulted to spot pairs. To see everything the connector can do, ask your AI: “List the TradingView tools you have access to.” And swap in the names of your own watchlists.

Official or open source: which one should you choose?

The official connector (this guide)

Two minutes, nothing to install, it works in Claude and ChatGPT alike. It reads prices (delayed during the beta), the screener, news, fundamentals and calendars, and it creates or edits your watchlists and alerts. It can't see your chart and doesn't write Pine Script.

The open-source bridge (the “Claude × TradingView Desktop” tab)

Longer to set up, but Claude drives your TradingView Desktop app: it writes your Pine Script indicators, compiles them and fixes its own errors — that's the setup behind our first two video tutorials. It doesn't place any orders either.

The two complement each other: nothing stops you from connecting both.

Good to know — this is a beta

TradingView says so itself: delayed market data and a still-limited tool list during the beta. For your scans, stick to H4 and the daily rather than the 15-minute.
About 100 tool calls per minute per user, and a possible daily cap during the beta: a scan of hundreds of assets across several timeframes can hit the limit. Start with a watchlist of a few dozen assets.
Some tools write to your account (creating, editing or deleting a watchlist or an alert — alerts are set on a price level): check what your AI is about to do before you confirm. No tool places orders.

Working from a terminal (Claude Code, Codex)?

One command is enough, then you authorize access in the browser (in Claude Code, type /mcp to start the authorization):

# Claude Code
claude mcp add --transport http mcp-tradingview https://mcp.tradingview.com/mcp

# Codex
codex mcp add tradingview --url https://mcp.tradingview.com/mcp

The complete guide

The 35 tools family by family, 10 copy-paste prompts, the beta’s limits, your account’s security, troubleshooting and the video tutorial: everything is gathered on one dedicated page.

Read the TradingView MCP guide →

Server provided by TradingView, in public beta: the tool list keeps changing — ask your AI to list the available TradingView tools to see today's version. Official documentation: tradingview.com/mcp/docs. Informational guide — not investment advice.

Before you start — have these ready

MetaTrader 5 installed and connected to your account · your 3 credentials given by your broker (account number, password, server name — in the account-opening email) · the Claude app on your computer (Claude Desktop). Budget ~15 minutes, no code to learn.

Unlike the TradingView bridge (read-only), this one can also place real orders. Make your first attempts on a demo account from your broker — zero risk, everything works the same.

1
Install Python
Go to python.org, click the big "Download" button, then open the downloaded file. Very important: on the installer's first screen, check the "Add Python to PATH" box at the bottom, before clicking "Install Now" — that's what makes the next step work (the famous PATH).
If a black window asks a question ending in [y/N], that means yes/no: type y then Enter to accept.
2
Install the MetaTrader bridge
Open the Start menu, type cmd and open "Command Prompt". In the black window, paste this then press Enter:
```
pip install metatrader-mcp-server
```
No cd to type, no folder to pick: the command installs itself in the right place no matter where you are. Once the text stops scrolling, it's done. If a "scripts … not on PATH" warning shows up, ignore it: all that matters is "Successfully installed … metatrader-mcp-server" — the guide launches the bridge via python -m, so that PATH doesn't come into play.
3
Let MetaTrader answer Claude
In MetaTrader 5: menu Tools → Options, tab Expert Advisors, check "Allow algorithmic trading", then OK. Leave MetaTrader open and connected to your account while you use Claude.
4
Open Claude’s settings file
In the Claude app: Settings (gear icon) → Developer tab → button "Edit Config". A file claude_desktop_config.json opens in Notepad — leave it open for step 5.
Can't find the button? Open File Explorer, paste this into the address bar then press Enter, and double-click the file:
```
%APPDATA%\Claude
```
5
Paste the setting and fill in your 3 details
In the open file: select all (Ctrl+A) and delete, paste the block below, replace the 3 words in CAPITALS with your MetaTrader credentials — keep the quotation marks — then save (Ctrl+S) and close.
```
{
  "mcpServers": {
    "metatrader": {
      "command": "python",
      "args": [
        "-m", "metatrader_mcp.cli",
        "--login",     "YOUR_ACCOUNT_NUMBER",
        "--password",  "YOUR_PASSWORD",
        "--server",    "YOUR_SERVER",
        "--transport", "stdio"
      ]
    }
  }
}
```
Your 3 credentials (account number, password, server name) are in your broker's account-opening email. Filled-in example: "--login", "12345678" · "--server", "MetaQuotes-Demo". If Claude says python isn't recognized, replace "python" with "py".
6
Restart Claude
Completely close the Claude app — also from the small icon near the clock (right click → Quit) — then relaunch it. The bridge is connected . On your first MetaTrader question, Claude will ask permission to use the tool: click "Allow".

Once connected — 5 prompts to try

1.“Show me my MetaTrader account balance and my open positions.”
2.“What is the current price of EUR/USD and gold (XAU/USD)?”
3.“Summarize my exposure: number of positions, direction (long/short), and my current profit or loss.”
4.“Analyze my account statistics: win rate, R:R ratio, Sharpe, maximum drawdown… and tell me my weak points.”
5.“On my demo account, prepare a 0.01 lot buy order on EUR/USD and explain what you are about to do before executing it.”

Security — your password sits in plain text in the file

Never share this settings file and never post it anywhere. Two healthy habits: use the "investor" (read-only) password if you only want to look, and always start on a demo account. And never paste your password into a conversation: it only belongs in the settings file.

If something goes wrong

"pip is not recognized" — Python never joined the PATH. Reinstall it (step 1), making sure to check "Add Python to PATH", then open a fresh black window.
Claude does not see MetaTrader after restarting — Almost always a missing quote or comma in the settings file: redo step 5 by pasting the clean block again, and make sure you fully quit and relaunched the app entirely (step 6).
"Could not attach" / "Server disconnected" — "Could not attach": Claude can't find the command — check "command": "python" (or "py"). "Server disconnected": the bridge started then stopped — MetaTrader must be open and connected, the checkbox from step 3 checked, and your 3 credentials exact (server name down to the last character).
Credentials rejected — Double-check the trio account / password / server — the server name must be exact, capitals and hyphen included, exactly as written in your broker's email.

Already set up with Claude Code (the Claude × TradingView guide)?

Claude Code does not read the app's file: it has its own config. A single command in its terminal (replace your 3 details):

claude mcp add metatrader -- python -m metatrader_mcp.cli --login YOUR_ACCOUNT_NUMBER --password YOUR_PASSWORD --server YOUR_SERVER --transport stdio

Express recap — keep this handy

# 1. Python: python.org -> Download -> check "Add Python to PATH"
# 2. The bridge (Command Prompt):
pip install metatrader-mcp-server
# 3. MetaTrader 5: Tools -> Options -> Expert Advisors
#    -> check "Allow algorithmic trading"
# 4. Claude: Settings -> Developer -> Edit Config
# 5. Paste the JSON block + your 3 credentials (account / password / server)
# 6. Quit Claude completely, then relaunch

Based on the open-source project ariadng/metatrader-mcp-server (MIT license). Informational guide — not investment advice. Trading carries a risk of loss: always test on a demo account before a real one.

Before you start — have these ready

Unlike the TradingView bridge (read-only), this one can also place real orders. Make your first attempts on a demo account from your broker — zero risk, everything works the same.

On Mac: a mandatory detour

MetaTrader's technical building block only exists for Windows: on Mac, the bridge doesn't install directly. The simplest fix: install Parallels Desktop, put Windows on it, then go through the 6 steps below inside that Windows — including MetaTrader 5 and the Claude app, installed inside it. Once there, everything works just like on a real PC.

1
Install Python
Go to python.org, click the big "Download" button, then open the downloaded file. Very important: on the installer's first screen, check the "Add Python to PATH" box at the bottom, before clicking "Install Now" — that's what makes the next step work (the famous PATH).
If a black window asks a question ending in [y/N], that means yes/no: type y then Enter to accept.
2
Install the MetaTrader bridge
Open the Start menu, type cmd and open "Command Prompt". In the black window, paste this then press Enter:
```
pip install metatrader-mcp-server
```
No cd to type, no folder to pick: the command installs itself in the right place no matter where you are. Once the text stops scrolling, it's done. If a "scripts … not on PATH" warning shows up, ignore it: all that matters is "Successfully installed … metatrader-mcp-server" — the guide launches the bridge via python -m, so that PATH doesn't come into play.
3
Let MetaTrader answer Claude
In MetaTrader 5: menu Tools → Options, tab Expert Advisors, check "Allow algorithmic trading", then OK. Leave MetaTrader open and connected to your account while you use Claude.
4
Open Claude’s settings file
In the Claude app: Settings (gear icon) → Developer tab → button "Edit Config". A file claude_desktop_config.json opens in Notepad — leave it open for step 5.
Can't find the button? Open File Explorer, paste this into the address bar then press Enter, and double-click the file:
```
%APPDATA%\Claude
```
5
Paste the setting and fill in your 3 details
In the open file: select all (Ctrl+A) and delete, paste the block below, replace the 3 words in CAPITALS with your MetaTrader credentials — keep the quotation marks — then save (Ctrl+S) and close.
```
{
  "mcpServers": {
    "metatrader": {
      "command": "python",
      "args": [
        "-m", "metatrader_mcp.cli",
        "--login",     "YOUR_ACCOUNT_NUMBER",
        "--password",  "YOUR_PASSWORD",
        "--server",    "YOUR_SERVER",
        "--transport", "stdio"
      ]
    }
  }
}
```
Your 3 credentials (account number, password, server name) are in your broker's account-opening email. Filled-in example: "--login", "12345678" · "--server", "MetaQuotes-Demo". If Claude says python isn't recognized, replace "python" with "py".
6
Restart Claude
Completely close the Claude app — also from the small icon near the clock (right click → Quit) — then relaunch it. The bridge is connected . On your first MetaTrader question, Claude will ask permission to use the tool: click "Allow".

Once connected — 5 prompts to try

1.“Show me my MetaTrader account balance and my open positions.”
2.“What is the current price of EUR/USD and gold (XAU/USD)?”
3.“Summarize my exposure: number of positions, direction (long/short), and my current profit or loss.”
4.“Analyze my account statistics: win rate, R:R ratio, Sharpe, maximum drawdown… and tell me my weak points.”
5.“On my demo account, prepare a 0.01 lot buy order on EUR/USD and explain what you are about to do before executing it.”

Security — your password sits in plain text in the file

If something goes wrong

"pip is not recognized" — Python never joined the PATH. Reinstall it (step 1), making sure to check "Add Python to PATH", then open a fresh black window.
Claude does not see MetaTrader after restarting — Almost always a missing quote or comma in the settings file: redo step 5 by pasting the clean block again, and make sure you fully quit and relaunched the app entirely (step 6).
"Could not attach" / "Server disconnected" — "Could not attach": Claude can't find the command — check "command": "python" (or "py"). "Server disconnected": the bridge started then stopped — MetaTrader must be open and connected, the checkbox from step 3 checked, and your 3 credentials exact (server name down to the last character).
Credentials rejected — Double-check the trio account / password / server — the server name must be exact, capitals and hyphen included, exactly as written in your broker's email.

Already set up with Claude Code (the Claude × TradingView guide)?

Claude Code does not read the app's file: it has its own config. A single command in its terminal (replace your 3 details):

claude mcp add metatrader -- python -m metatrader_mcp.cli --login YOUR_ACCOUNT_NUMBER --password YOUR_PASSWORD --server YOUR_SERVER --transport stdio

Express recap — keep this handy

# 1. Python: python.org -> Download -> check "Add Python to PATH"
# 2. The bridge (Command Prompt):
pip install metatrader-mcp-server
# 3. MetaTrader 5: Tools -> Options -> Expert Advisors
#    -> check "Allow algorithmic trading"
# 4. Claude: Settings -> Developer -> Edit Config
# 5. Paste the JSON block + your 3 credentials (account / password / server)
# 6. Quit Claude completely, then relaunch

Before you start

Use Windows PowerShell (not Git Bash, not the x86 version). TradingView comes as an MSIX (Microsoft Store) package — that's normal. Never work from C:\Windows\System32 or C:\. After each install: close PowerShell and open a fresh window. Read-only setup — Claude reads your chart, it never places an order. Open-source repo: tradesdontlie/tradingview-mcp.

1
Check what is already installed
```
git --version
node --version
winget --version
```
If any of them returns "not recognized", go to step 2 to install it.
2
Install what is missing
```
# If Git is missing
winget install Git.Git

# If Node.js is missing
winget install OpenJS.NodeJS.LTS
```
Accept the "run as administrator" window if it appears. Then close/reopen PowerShell and check with node --version.
3
Allow scripts (for npm)
```
Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned
```
If asked to confirm, type Y then Enter. Check with Get-ExecutionPolicy -Scope CurrentUser → it must display RemoteSigned. Fallback: otherwise replace npm with npm.cmd in every command.
4
Install TradingView Desktop
```
winget install TradingView.TradingViewDesktop
```
Important: then launch TradingView once normally, log in, and open a real chart (e.g. BTCUSD). This first launch finalizes the installation. Then close it.

Install Claude Code

irm https://claude.ai/install.ps1 | iex

Close/reopen PowerShell, then: claude --version. If "claude is not recognized" (PATH not set):

Test-Path "$env:USERPROFILE\.local\bin\claude.exe"
# if True, apply the permanent PATH fix:
[Environment]::SetEnvironmentVariable("Path", $env:Path + ";$env:USERPROFILE\.local\bin", "User")

Close/reopen PowerShell and test again. The PATH change is permanent — one time only.

6
Create the working folder and launch Claude Code
```
cd $env:USERPROFILE
mkdir tradingview-claude
cd $env:USERPROFILE\tradingview-claude
claude
```
On screen: "Is this a project you created or one you trust?" → choose 1 (Yes, I trust this folder), only if the folder shown really is tradingview-claude. On first launch, log in to your Claude account (Pro or Max) in the browser.
7
Install the MCP server (manual method, recommended)
Exit Claude Code (/exit) to get back to PowerShell, then:
```
cd $env:USERPROFILE\tradingview-claude
git clone https://github.com/tradesdontlie/tradingview-mcp.git
cd tradingview-mcp
npm install
```
Register the server in Claude Code (the path builds itself):
```
claude mcp add tradingview -- node "$($PWD.Path)\src\server.js"
```
You should see: "Added stdio MCP server tradingview… to local config". It is registered.
8
Launch TradingView with the debug port (key step)
1. Close TradingView completely (window AND system-tray icon, right click → Quit). 2. Locate it and launch it with port 9222:
```
$tv = (Get-AppxPackage *TradingView*).InstallLocation
Start-Process "$tv\TradingView.exe" -ArgumentList "--remote-debugging-port=9222"
```
3. Reopen a real chart. 4. Check that the port responds:
```
(Invoke-WebRequest http://localhost:9222/json/version -UseBasicParsing).Content
```
You should see: a JSON block (Browser, webSocketDebuggerUrl…) → the port is open, go to step 9. Tip: keep the 2 lines from point 2 in a start-tradingview.ps1 file on your desktop (right click → Run with PowerShell).
9
Check the connection in Claude Code
```
cd $env:USERPROFILE\tradingview-claude
claude
```
At the prompt, type: "Run tv_health_check and tell me if TradingView is connected." You should see: cdp_connected: true, api_available: true, your symbol and your timeframe. If api_available: false → open a real chart and run the health check again.

Express recap (once everything is installed)

# 1. Close TradingView, then relaunch it with the debug port
$tv = (Get-AppxPackage *TradingView*).InstallLocation
Start-Process "$tv\TradingView.exe" -ArgumentList "--remote-debugging-port=9222"

# 2. Open a real chart in TradingView

# 3. Check the port responds (must return a JSON block)
(Invoke-WebRequest http://localhost:9222/json/version -UseBasicParsing).Content

# 4. Launch Claude Code in the project folder
cd $env:USERPROFILE\tradingview-claude
claude
# then in Claude: "Run tv_health_check"
# -> cdp_connected: true, api_available: true

Before you start

Use the Terminal (Applications → Utilities — zsh, the macOS default shell). First opening of an app blocked by Gatekeeper? Right click → Open. Never work from a system folder. After each install: close Terminal and open a fresh window. Read-only setup — Claude reads your chart, it never places an order. Open-source repo: tradesdontlie/tradingview-mcp.

1
Check what is already installed
```
git --version
node --version
brew --version
```
If any of them returns "command not found", go to step 2 to install it. (macOS may offer to install its developer tools on the first git --version — accept, that is enough.)
2
Install Homebrew and what is missing
```
# Homebrew (the macOS package manager), if missing:
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Then:
brew install git node
```
On Apple Silicon, Homebrew prints two "Next steps" commands at the end of the install to add itself to the PATH — run them. Then close/reopen Terminal and check with node --version.
3
Install TradingView Desktop
```
brew install --cask tradingview
```
Important: then launch TradingView once normally (if macOS blocks the first opening — Gatekeeper —: right click on the app → Open), log in, and open a real chart (e.g. BTCUSD). This first launch finalizes the installation. Then quit it completely (⌘Q).
4
Install Claude Code
```
curl -fsSL https://claude.ai/install.sh | bash
```
Close/reopen Terminal, then: claude --version. If "command not found" (PATH not set):
```
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc
```
Then test again. The line is permanent — one time only.
5
Create the working folder and launch Claude Code
```
cd ~
mkdir tradingview-claude
cd ~/tradingview-claude
claude
```
On screen: "Is this a project you created or one you trust?" → choose 1 (Yes, I trust this folder), only if the folder shown really is tradingview-claude. On first launch, log in to your Claude account (Pro or Max) in the browser.
6
Install the MCP server (manual method, recommended)
Exit Claude Code (/exit) to get back to Terminal, then:
```
cd ~/tradingview-claude
git clone https://github.com/tradesdontlie/tradingview-mcp.git
cd tradingview-mcp
npm install
```
Register the server in Claude Code (the path builds itself):
```
claude mcp add tradingview -- node "$PWD/src/server.js"
```
You should see: "Added stdio MCP server tradingview… to local config". It is registered.
7
Launch TradingView with the debug port (key step)
1. Quit TradingView completely (⌘Q — check the menu-bar icon too). 2. Relaunch it with port 9222:
```
open -a TradingView --args --remote-debugging-port=9222
```
3. Reopen a real chart. 4. Check that the port responds:
```
curl -s http://localhost:9222/json/version
```
You should see: a JSON block (Browser, webSocketDebuggerUrl…) → the port is open, go to step 8. Tip: keep the command in a start-tradingview.command file on your desktop (once: chmod +x on it) → double-click to launch.
8
Check the connection in Claude Code
```
cd ~/tradingview-claude
claude
```
At the prompt, type: "Run tv_health_check and tell me if TradingView is connected." You should see: cdp_connected: true, api_available: true, your symbol and your timeframe. If api_available: false → open a real chart and run the health check again.

Express recap (once everything is installed)

# 1. Quit TradingView (Cmd+Q), then relaunch it with the debug port
open -a TradingView --args --remote-debugging-port=9222

# 2. Open a real chart in TradingView

# 3. Check the port responds (must return a JSON block)
curl -s http://localhost:9222/json/version

# 4. Launch Claude Code in the project folder
cd ~/tradingview-claude
claude
# then in Claude: "Run tv_health_check"
# -> cdp_connected: true, api_available: true

The tutorials — connecting Claude, then putting it to work

New · The tutorial from September 23rd: the official MCP installed in ChatGPT and then in Claude, a tour of its 35 tools, and a scan that saves its watchlist straight to TradingView — the written version is just above, in the guide's first tab.

The tutorial from July 27th: the full connection in 13 minutes, step by step, without writing a single line of code — the written version is just above, in the "Claude × TradingView Desktop" tab.

The tutorial from August 29th: Claude scans hundreds of assets on OKX and generates your TradingView watchlists — so you stop hunting for trades by hand.

Watch every video in English — Our videos are recorded in French, but YouTube dubs them all into English automatically: as soon as your YouTube language is set to English, the English audio track plays by default. On youtube.com, click your profile picture → "Language" → English — the mobile app simply follows your phone's language. You can also switch any video by hand, right in the player: Settings → "Audio track" → English.

Definition

AI trading: what are we talking about?

AI trading — sometimes written "trading AI" — describes the use of artificial intelligence to analyze markets, generate trade ideas, or place orders. The problem is that the same term gets used to sell everything and its opposite: a trading bot for a few dozen euros a month, an app promising a "95% win rate," the statistical model behind an investment fund, a conversational assistant. None of these have much in common: not the technology, not who is deciding, not the risk.

The usual breakdown covers algorithmic trading tools, signal generators, conversational generative AI, and robo-advisors. The classification that follows is ours, built on two simple questions: who decides, who executes? Signal generators are not a separate family here: they are bots or models that display an idea instead of executing it.

1. Rule-based bots (automated trading)

Examples: exchange grid bots and DCA bots, Expert Advisors on MetaTrader, KryllOS, 3Commas, Cryptohopper

Who decides? you, by writing the rules

Who executes? the bot, on your account — built into the exchange (grid, DCA), installed in your MetaTrader terminal (Expert Advisor), or connected via an API key (Kryll, 3Commas, Cryptohopper)

Framework: the bot itself is not regulated; the platform that receives its orders must be regulated

"AI" is often just marketing here: a grid bot does not think.

2. Quantitative models and machine learning

Examples: statistical arbitrage, predictive models (regression, time series, neural networks), high-frequency trading

Who decides? the model, based on probabilities estimated from the past

Who executes? the machine, with no human hand — anywhere from a microsecond to a few days

Framework: the world of funds and professional trading desks — expensive data, infrastructure, and teams

Here, the model decides and executes on its own — and high frequency is more about speed than AI. A model that wins is not rented out by subscription: whoever owns it keeps it. What an individual trader can actually buy are scanners and pattern detectors (see the comparison).

3. Generative AI as copilot (LLM)

Examples: ChatGPT, Claude, Gemini, Perplexity, Mistral

Who decides? you

Who executes? no one — unless you connect the AI to your platform via an MCP server, and even then it only acts on your request

Framework: an LLM is neither a financial advisor nor regulated; unconnected, it does not know today's price; it can be confidently wrong

This is the family we actually tested: coding an indicator, scanning hundreds of assets, setting alerts. What you gain is time, never a decision.

4. Regulated robo-advisors

Examples: robo-advisors from licensed fintechs and the automated managed portfolios offered by some banks or online brokers

Who decides? the allocation algorithm, based on your profile, if you have signed a management mandate; you, if the offer is limited to advice you approve

Who executes? the asset management firm acting under mandate, or the broker hosting your account

Framework: a licensed asset management firm, or a registered investment advisor; a mediator you can turn to in case of a dispute

This is not trading — it is long-term managed saving. It is the only one of the four where a licensed professional is accountable — though still with no guarantee of performance.

One last confusion to clear up: "trading AI" has nothing to do with "trading with AI." Buying Nvidia, Microsoft, or a themed ETF because you believe in the sector is a bet on a theme. Working with an LLM or a bot is a method. This page is about the second one.

The tools

AI trading tools in 2026: the comparison

What is the best AI for trading? There is not one — there is an AI for each task. For coding an indicator or summarizing a report, an LLM. For placing orders on a set schedule, a rule-based bot. For spotting a chart pattern across thousands of instruments, a screener. For each tool, the comparison says what it does and what "AI" means for it, whether it can place orders, whether it shows real-time prices, and what it costs. The badges under each name classify by use (LLM copilot, charts, platform, bot, signals), not by the four families above. Two honest caveats: prices change, so verify before you pay, and we have only tested one tool ourselves, Claude — on video for TradingView, in a written guide for MetaTrader 5 (the MCP setup on this page, for its part, assumes two paid subscriptions). For every other tool, the comparison sticks to what the publisher states: no line is an opinion, let alone a recommendation. Robo-advisors are not included: that is managed saving, not trading.

Prices checked on 18 septembre 2026

Tool	What it does	Where AI fits in	Orders	Real-time	Price	Free	Our experience
ChatGPT (OpenAI)	Conversational assistant: text, web search, files, Python code, Deep Research. No broker, no quotes.	GPT LLM (5.6 Luna on Free and Go, 5.6 Sol from Plus, GPT-6 Pro on the Pro plan), with web search, code, and browser.	No — no broker or exchange connected	No — web search, no quotes	from 8 €/mo (Go); Plus 23 €; Pro from 103 €	Yes (with ads)	—
Claude (Anthropic)	Conversational assistant: text, files, web search, code, MCP connectors (this is the one used on this page, connected to TradingView and MetaTrader 5).	Claude LLM (Sonnet and Haiku free; Fable and Opus paid), with web search, code, and MCP connectors.	Via broker — yes on MetaTrader 5 or a connected broker, for the order you dictate; no on TradingView (neither the open-source bridge nor the official connector places orders)	No — no quotes natively; TradingView's once the MCP bridge is in place	from 17 $/mo billed annually (20 $ monthly); Max 100 to 200 $, excl. tax	Yes	Yes — on video for TradingView; written guide for MetaTrader 5
Gemini (Google)	Google assistant: answers, files, Deep Research backed by Google Search. No broker; prices live separately in Google Finance.	Gemini LLM (3.6 Flash free, 3 Pro on Pro and Ultra, Deep Think on Ultra), grounded in Google Search.	No — no broker app connected	No — Google search, no quotes	from 4.99 €/mo (Plus); Pro 21.99 €; Ultra from 99.99 €	Yes	—
Perplexity (Perplexity AI)	Sourced answer engine plus a Finance section (prices, screener, watchlist, read-only portfolio). Research, not trading.	Orchestrated third-party LLMs (GPT, Gemini, Claude…) plus its own Sonar model; the Finance section reads an FMP data feed.	No — read-only Plaid link, United States and Canada	Yes — on-demand prices in the Finance section, delay not guaranteed	from 17 $/mo billed annually (20 $ monthly); Max 200 $; 22 € via the App Store	Yes	—
Mistral Vibe (Mistral AI)	The Paris-based publisher's former Le Chat: Work mode (agent, web, connectors) and Code mode. No broker, no market data.	Mistral Medium 3.5 LLM, with web search, deep research, and MCP connectors. No market model.	No — mail, calendar, Slack, GitHub connectors	No — web search, no quotes	from 14.99 €/mo incl. tax (Pro); Team 24.99 €/user	Yes	—
Grok (xAI)	Assistant built into X: answers, live web and X search, images, code, Grok Bot agents. No trading.	Grok 4.6 LLM, with tool-equipped web and X search, Expert and multi-agent modes.	No — no broker	No — live web and X, no quotes	30 $/mo (SuperGrok); Plus 100 $; via X Premium from 8 € excl. tax	Yes	—
DeepSeek (DeepSeek)	Free chat and app from the Chinese publisher: answers, DeepThink reasoning, web search. No broker.	In-house DeepSeek-V4.1 LLM (open weights), with web search. No market data.	No — no broker connection	No — web search, no quotes	free (chat and app); API from 0.30 $ per million tokens	Yes, no subscription	—
TradingView (TradingView)	Charts, screeners, alerts, Pine Script; orders via the Trading Panel at more than 100 partner brokers. TradingView is not a broker.	4 tools in beta: AI Chart Copilot, natural-language AI Screener, SEC summaries, MCP server.	Via broker — Trading Panel, from the Basic plan	Depends on plan — crypto and forex free; stocks by exchange (Euronext 9.95 €)	from 12.95 €/mo billed annually (14.95 € monthly, Essential)	Basic	Yes — on video (Claude × TradingView, tutorials and replays from this page)
MetaTrader 5 (MetaQuotes)	Platform distributed by brokers: orders, MQL5 robots (Expert Advisors), Strategy Tester, Market, VPS. A software publisher, not a broker.	3 layers: rule-based MQL5 robots; ONNX inference; AI Assistant LLM and MCP (since 07/2026).	Yes — on your broker account; AI actions require confirmation	Yes — feed from the connected broker	free (software and AI Assistant, paid subscription announced); Market from 30 $	Yes	Written guide — open-source MCP bridge, verified before publishing
Lutessia (CentralCharts)	Technical analyses generated for more than 27,000 instruments (opinion, invalidation, targets), on demand or via scanner. No execution signals.	In-house rules and scoring engine, validated by humans. Neither an LLM nor machine learning.	No — no broker connection	No — closed candles; stocks delayed 15 min	incl. tax: 29.90 €/mo, 161.46 €/6 months, 287.04 €/yr	No	—
KryllOS (Cryptense, ex-Kryll)	Self-hosted open-source crypto bot from the French publisher: block-based strategies, backtesting, live trading via local API keys. Invitation-only access.	An LLM translates a description into strategy blocks; agents announced. Model undocumented.	Yes — Binance, Coinbase, Hyperliquid via API keys	Yes — feed from the connected exchange	free (software, no subscription); "AI tokens" not priced	Yes	—
3Commas (3Commas)	Crypto bots (DCA, Grid, Signal, SmartTrade) running an exchange account via API keys; AI Assistant and QuantPilot for configuration.	Rule-based bots; the AI Assistant is a Gemini LLM that configures DCA bots; QuantPilot runs agents on an unnamed model.	Yes — 9 exchanges via API keys; not in the free plan	Yes — feed from the connected exchange	from 15 $/mo billed annually (20 $ monthly); Pro 50 $; Expert 140 $	Yes (no real trading)	—
Cryptohopper (Cryptohopper)	Dutch cloud crypto bot: indicator strategies, marketplace, TradingView alerts, paper trading; orders via API keys; a 10, 5, or 2-minute decision cycle depending on the plan.	"Algorithm Intelligence": a score-based strategy selector, no LLM.	Yes — 11 exchanges, API keys with no withdrawal rights	Yes — feed from the connected exchange, read every cycle	from 24.16 $/mo billed annually (29 $ monthly); Hero 129 $	3-day trial	—
Capitalise.ai (Kraken)	No code required: a strategy typed in English became rules executed at the connected broker. Service discontinued since the Kraken acquisition.	Language processing that translates a sentence into deterministic rules. No prediction.	No — service discontinued (IBKR partnership ended in March 2026)	No — service discontinued	was free via partner brokers	Not applicable	—
Trade Ideas (Trade Ideas)	Scanner and charts for US stocks; Holly AI issues daily signals; orders via Brokerage Plus.	Holly runs more than 60 rule-based strategies re-backtested every night, on proprietary parameters. Neither an LLM nor a published model.	Via broker — Brokerage Plus: IBKR, E*TRADE, Alpaca	Depends on plan — free delayed 15 min, live on paid plans	from 89 $/mo billed annually (127 $ monthly); Premium 254 $ (Holly)	Yes (delayed data)	—
TrendSpider (TrendSpider)	Charts, pattern detection, scanners, backtests, bots; Sidekick assistant; orders via SignalStack.	Algorithmic pattern detection plus Sidekick, a chatbot built on third-party LLMs (Gemini, Claude, GPT).	Via broker — SignalStack (in-house), 33 brokers	Yes — US stocks, forex, crypto; futures optional	from 52.38 $/mo billed annually (82 $ monthly), intro rate then renews higher; Sidekick 49 $	No	—
Tickeron (Tickeron)	US platform of "AI Robots": intraday signals (60, 15, or 5-minute depending on the plan) on stocks, ETFs, and cryptos; the client copies the trades over to their broker.	"Financial Learning Models": proprietary machine learning on a simulated account.	No — manual copy; no broker connected	No — timestamped signals, no continuous quotes	from 90 $/mo; Expert 250 $ or 1,500 $/yr	Beginner	—

Situation observed on the publishers' official pages. Public prices excluding promotions, in the currency shown; an entry-level rate is marked as such; "incl. tax" only when the publisher states it. "Orders": "Yes" means the tool sends the order to your account itself. "Via broker" means the order goes to a connected third party (connector, trading panel, mandate). "No" means no execution at all, manual copying included. "Real-time" refers to market quotes, not an assistant's web updates; "Depends on plan" means live data depends on the plan or the exchange. "Our experience": "Yes" means the Captain has used it in a video on this page; "Written guide" means we documented it without filming it; "No" means the line is based on the publisher's documentation. None of these tools is an investment advisor or a regulated financial intermediary: when one of them places an order, it is the broker or exchange behind it that must be regulated. The performance figures publishers display are not repeated here.

The setup

What we actually plugged in

No magic involved. An open-source MCP server (Model Context Protocol) — in plain terms, a bridge that lets Claude drive another piece of software, here TradingView Desktop. Claude doesn't "enter" TradingView, it controls it: it opens the app, selects assets, and above all it writes Pine Script, the language behind TradingView indicators. You describe what you want in plain English. It executes. No need to be a developer, no need to understand what's happening under the hood.

It all started with a slightly provocative question, asked in the live session on June 23rd: does Claude trade better than you? The Captain's answer fits in one sentence: the real question isn't whether the AI will trade in your place, but how you can get better with its help. Two exploratory live sessions later, everything this setup makes possible — building a custom Pine Script indicator from a simple price action idea, building watchlists on your own technical criteria, setting alerts where TradingView doesn't offer any, or backtesting a strategy — now fits in the tutorial above: 13 minutes flat to reproduce the full connection, step by step. The full replays of both live sessions are still further down this page.

The short from July 3rd — the setup summed up in 42 seconds.

The demo

Pine Script without knowing how to code: what it lets you do

A week later, it's time to put it into practice: part 2 happens live, no safety net. The team loves the Choppiness Index — the "chop" —, an indicator that measures whether the market is going sideways or ripe to send a trend. Problem: TradingView doesn't offer any native alert for it. Until now, you needed real Pine Script skills to build one. In a few commands, Claude wrote the alert indicator. A compile error along the way? It fixed it itself. End result: an alert on crossing the 70 threshold, on H1, H4 and daily — on crossing only, to avoid spam. All of it live, in front of the members, hesitations and corrections included.

Here's the detail that changes everything: the Captain can't code. Zero programming knowledge, and his first indicator was born in a matter of moments. And that's only the start: building other custom indicators, generating then backtesting strategies without coding, building dashboards and watchlists on your own parameters. In the short from July 7th — shown here —, the Captain goes further: instead of combing through 200 charts by hand, Claude scans the market against his criteria in a single command, and surfaces a trade he would probably have missed on his own.

The short from July 7th — 200 charts scanned in 49 seconds.

The replay — the indicator coded live

Part 2 — the live session from June 30th: the Pine Script indicator coded live, then the market analysis (46 min).

Got a question? Ask the Captain directly

The community channel is plugged live into this page — the Captain and the members talk AI, bots and automation all day long, auto-translated into English in real time. Read freely; to ask your question, log in and write it in English: it lands in the Discord, translated into French for the community.

#ai-and-trading · live from the Discord

AI, bots and automation — the community's dedicated channel, auto-translated from French.

1,073 online10,466 members

Pariston29 Sept 20:17

thanks a lot man

₿ CryptoSkud ₿29 Sept 20:19

Happy to help sorry we didn't get all the way through. Give him the explanation here that we did together that way he'll know where you're at: https://discord.com/channels/739401711000748062/1496835610709594132/1554545078612004894

Pariston29 Sept 20:19

ok man thanks a lot

₿ CryptoSkud ₿29 Sept 21:45

Keep me posted if you pull it off 💪

₿ CryptoSkud ₿30 Sept 11:49

So?! Did it work out?!

*Solaris* Tortue8 Oct 22:17

Hey @₿ CryptoSkud ₿ I just sent you a DM. 🙂

₿ CryptoSkud ₿8 Oct 22:23

I don't think it was me, I didn't get anything

*Solaris* Tortue8 Oct 22:24

Yeah, but you have to send a friend request to contact you. Which I did too

₿ CryptoSkud ₿8 Oct 22:26

I'm telling you I got nothing, I'm sending you a friend request, send me your dm and I'll check it out tomorrow

*Solaris* Tortue8 Oct 22:27

Resent

farfarod11 Oct 12:51

hey everyone does this script work on ninja trader too??? have a good sunday

Le captain11 Oct 14:03

which script?

farfarod11 Oct 15:02

@Le captain sorry I reposted my question in the right channel. Go ahead and ask your question. I'm trying to find out if the captain's chop indicator can be integrated into ninja trader. Because you have to build it from scratch

Alex14 Oct 15:36

Hey everyone! Alexy here, been a dev for over 11 years now, and I'm starting to think that trading and AI (already through automation) are gonna become more and more doable. The breakthrough is probably coming soon if it's not already here. Personally, I'm a total noob at trading, just want to see where to start, and if anyone's already started working on AI auto-trading projects? (Though I'm guessing you didn't wait for me to get going on that 😂) Looking forward to chatting about it if people are keen to seriously dive into this :)

Carlito15 Oct 20:24

Hey, my LWMA / SMMA trading bot isn't taking any trades :/ any idea where that could be coming from? Thanks 🙂

₿ CryptoSkud ₿15 Oct 20:26

Hey! Probably the start time 21 & end time 21

₿ CryptoSkud ₿15 Oct 20:26

Disable the time filter out of curiosity

₿ CryptoSkud ₿15 Oct 20:27

And of course verify the backtests with the parameters, if it takes nothing in backtest, it won't take anything live

Carlito15 Oct 20:31

it's performing well in backtest. FYI the first time I created the bot there were curves with arrows and then overnight nothing, even though I rebuilt the bot from scratch

C-Lakata15 Oct 21:52

Hey, probably filtered by max ADX at 17 …?

Carlito15 Oct 22:33

Hey, yeah I'm gonna check it out

Carlito15 Oct 22:45

Same old, same old :/

BBS-Starter16 Oct 08:50

send your logs during the backtest

BBS-Starter16 Oct 08:50

just to me if you want

Arno16 Oct 08:57

Is that demo account still active? 14 day lifespan?

matmahvv18 Oct 16:33

Hey me too I have the same problem

BBS-Starter18 Oct 18:09

Like we just saw with matmahvv, right now the strat isn't taking any trades on gold because nothing's happening on gold

BBS-Starter18 Oct 18:09

the filter is doing its job and blocking trades that would be losers

TomusTheOne21 Oct 01:56

@Carlito I can confirm, I've been testing it for almost 2 weeks since October 8th on gold and haven't gotten any trades either, but it's working really well on the footsie and eur/usd. Today I hit a bug (5 closes for no reason) after adjusting partial tp on eth this weekend (and I added ibex and gbp/jpy yesterday so zero trades on those for now) but eth is good and footsie eur/usd are really solid!

TomusTheOne21 Oct 19:07

@Carlito 1st trade on gold went out an hour ago (bot launch on the 8th)

You can read freely — to join in, log in.

Transparency

The limits, no filter

We're not here to sell you a dream, that's not how this place works. Three things to know before you get started.

The cost

You need a paid Claude subscription and a paid TradingView plan, and the tokens burn fast.

The setup

It can be a real pain — the Captain struggled with it, a colleague didn't, go figure.

No miracle strategy

The AI doesn't have a hidden holy grail wired into its circuits. An excellent assistant, not an oracle.

"It will never replace your judgment as a trader, only the time you used to waste on repetitive tasks."
The Captain, live session of June 30th

And what the AI produces can be checked: a backtest is only ever a test on the past, never a promise about the future.

Budget

How much AI trading costs, line by line

We did not find a single fully costed setup in the results we checked on September 18, 2026. Here is ours, checked the same day on the official sites, in dollars and euros with no conversion. The dollar prices are the ones the publishers display, plus tax where applicable (Claude states it: "applicable tax not included"); TradingView's are shown as a "special price" and may be increased by tax depending on your location. They change: verify before you pay.

1. Testing it out, like in the tutorials and the guide

Claude Pro: 17 $/mo billed annually (20 $ monthly)
TradingView Essential: 12.95 €/mo billed annually (14.95 € monthly) — 5 indicators per chart, 20 alerts
MetaTrader 5 and its demo account (written guide on this page), open-source MCP server: 0 €

Total: 17 to 20 $ + 12.95 to 14.95 € per month

No real-time NYSE, Nasdaq, or Euronext without a paid package: the free US feed is limited to the Cboe BZX feed.

2. Running it every day

Claude Max 5x: 100 $/mo, monthly billing only
TradingView Plus: 29.95 €/mo billed annually (34.95 € monthly) — 10 indicators, 100 alerts
Euronext real-time package: 9.95 €/mo
VPS to keep the MetaTrader 5 bridge running: an entry-level VPS from a European host (checked at OVHcloud: 4.57 € incl. tax/mo, the only tax-included price on this card) or the MetaQuotes VPS, 10 to 15 $/mo

Total: 100 $ + 44.50 to 49.50 € per month with the European VPS (110 to 115 $ + 40 to 45 € with the MetaQuotes VPS)

This is the tier where the TradingView subscription stops being the constraint.

3. A rule-based bot on an exchange

Exchange-native grid bot: 0 € subscription — you pay fees on every order (typical European fee grid: 0.08% maker, 0.10% taker on spot)
Third-party bot: 3Commas from 15 $/mo billed annually, Cryptohopper from 24.16 $/mo billed annually
Expert Advisor from the MetaTrader 5 Market: 79 to 1,999 $ as a one-time purchase, per the storefront on September 18, 2026, plus 10 to 15 $/mo for a VPS

Total: 0 € to 25 $ in subscription per month, plus 10 to 15 $ for a VPS if you run an Expert Advisor, plus fees

For reference: 20 round trips a day at 0.10% on 500 € comes to 20 € in fees per day, enough to swamp every subscription. And the real line item is not the subscription: a bot puts your capital on the line with every order, not just its fees.

The hidden cost is context. On a Pro or Max subscription, there is no surprise bill: once the five-hour window is used up, Claude stops and tells you, and extra credits only kick in if you request them yourself. But an MCP server that dumps out candle history fills that window fast — a 10 KB web page already comes to about 2,500 tokens — and Claude Code, included in the subscription, draws from the same pool. That is what "the tokens run out fast" means. The API, on the other hand, is billed by usage: that is not the setup on this page. The only free line item in the table is the MetaTrader 5 demo account — that is where you start, never with real money.

The verdict

What the AI won't do in your place

The last video on this page isn't about AI, and that's on purpose. A NASDAQ trade, pure method, in three steps: spotting a key zone — here a supply zone —, reading the volume of the last rally on the Volume Profile, then waiting for the breakout to fail. Take profit set on the next key zone, counter-trend and fully owned, and a line that sums it all up: "no need to shoot for the moon."

Look closely: no AI is involved. Reading the market, the plan, the patience, the execution discipline — that's all brought by the trader. That's the real verdict of the "AI vs trading" match: there isn't one. The AI absorbs the repetitive mechanics, the trader keeps the decision. And that part — the only one that really matters — is something you learn.

A NASDAQ trade in 3 steps — 100% human judgment.

The series continues

The latest AI videos

We keep exploring AI applied to trading, live session after live session. New videos in the series are added here automatically, as soon as they're published on the YouTube channel.

Je code un indicateur TradingView avec ChatGPT… sans savoir coder — September 30, 2026

ALERTE NASDAQ : les taux US s’envolent, la bulle IA est-elle en train de lâcher ? — September 27, 2026

ChatGPT Astra me génère des watchlist sur Tradingview — September 21, 2026

Chat GPT Astra X Tradingview: nouveau MCP OFFICIEL — September 17, 2026

Automatiser sa Stratégie de Trading : Comment Passer du Manuel à l'Automatique — September 10, 2026

MCP MetaTrader 5 : Claude passe tes ordres ! — September 8, 2026

Le secret pour trouver des opportunités de trading en 60s #ia #crypto #trading — September 8, 2026

Les Prompts qui Changent Tout pour Trader avec l'IA — September 4, 2026

Concevoir un Bot de Trading : Comment Ça Marche Vraiment ?! Avec FX et BBS — September 3, 2026

Contrôler MT5 avec Claude en moins de 5 minutes (Guide MCP complet) — September 1, 2026

Trader avec l'IA en 2026 : Ce qui Marche Vraiment — August 31, 2026

The deep dive

AI Trading: the complete guide

The Claude × TradingView demo is the visible tip. For the substance, the complete guide from our free trading course now lives here: what AI trading really covers, how machine learning changes analysis, the traps to know — and the automated BTC/ETH strategy built with the community.

Expand the complete guide (≈ 23 min read)

AI trading is defined as the use of artificial intelligence in the process of analyzing market data, either to generate investment ideas or to act directly on our behalf. The use of AI in the markets has revolutionized the financial sector and the trading world by making it more efficient. If AI trading has grown so popular in recent years, and so many traders and investors are adopting it, there’s a good reason for it: the Results!

AI trading generally involves the use of algorithms and machine-learning techniques to analyze vast amounts of data and identify patterns and trends in the market.

Thanks to a computer program, the risk of errors between intention and execution — what we also call human error — can be considerably reduced!

Precision is another undeniable asset of trading assisted by artificial intelligence!

In short, you get the idea: going without artificial intelligence in the markets today is a shame at the very least, if not a considerable amount of money left on the table.

That’s why the Captain Trading community has begun developing automated systems dedicated to decision-making in the financial markets.

So in this guide on AI trading, alongside an overview of the topic, we’ll walk you through the full logic of an automated BTC and ETH strategy, designed from A to Z by a member of our community!

Heads-up: if you’re mainly interested in our automated strategy, please note that it relies on perpetual and inverse perpetual contracts; we run it on OKX. Check whether derivatives are available from your country before you start. Depending on the platform you choose, your results will differ: this is a strategy that borders on high-frequency trading, where the smallest differences in market movements can have a considerable impact!

AI Trading: the Guide!

AI Trading: Definition

AI trading, also known as algorithmic trading or algo trading. It’s a method of executing transactions on the financial markets using computer algorithms. These algorithms analyze larger or smaller amounts of data — such as historical price movements, market trends, and economic indicators — in order to identify patterns and make trading decisions.

Recently, the use of artificial intelligence in the markets has gained ground thanks to its ability to analyze large amounts of data quickly and accurately, which lets it identify patterns and make decisions faster than any human could ever hope to.

Trading-focused AIs have evolved considerably over the years, using increasingly sophisticated machine-learning algorithms. The use of AI in trading has quite simply allowed traders to make better decisions by analyzing large amounts of data quickly and accurately.

On top of that, AI has enabled some traders to automate their trading strategies, allowing them to seize market opportunities 24 hours a day, 7 days a week. And I’ve already given you a taste of this in the first part!

Of course, artificial intelligence is used on traditional markets as well as in cryptocurrency trading.

AI Trading: the Main Uses and Technologies

machine learning,
natural language processing
massive-data analysis: big data
Automatic order execution

Machine-learning algorithms are used to analyze large amounts of data in order to identify patterns and make longer-term trading decisions.

Natural language processing makes it possible to analyze news articles and other information sources in order to identify market trends and opportunities.

Massive-data analysis (big data) is used to analyze large amounts of data in order to identify patterns and trends in the market.

Beyond these 3 main aspects, AI trading platforms also use advanced algorithms to execute transactions automatically. These algorithms are designed to capitalize on market opportunities the moment they arise, letting traders make better decisions without delay and thereby increase their profitability.

Overall, AI trading is a fast-moving field that offers traders a whole range of advantages. By using advanced algorithms and technologies, traders can analyze large amounts of data quickly and accurately, identify market trends and opportunities, and automate their trading strategies to capitalize on market opportunities 24 hours a day, 7 days a week.

AI Trading: the Advantages

There are many reasons why investors and traders may be drawn to artificial intelligence — we’ve already mentioned a few, but it can go much further! Here are some of the specific benefits of using artificial intelligence in trading:

Reduced research time
Automation
Better forecasts and reporting
Cost reduction
Scalable technology
No emotions
Fast trade execution
Effective backtesting

All in all, AI trading — trading assisted by artificial intelligence — brings tremendous added value to the financial markets!

Algo Trading: the Different Types of Algorithmic Trading

Quantitative Analysis

Quantitative analysis can be applied to both technical and fundamental analysis. It’s a popular approach to algorithmic trading. It involves using mathematical models and statistical techniques to identify patterns and trends in market data. A trader can use these models to develop trading strategies that exploit market inefficiencies and other opportunities.

For example, regression analysis is a very common quantitative-analysis technique. In fact, if you went to business school, the term should ring a bell — especially if I say “linear regression,” right?!

Regression analysis is a statistical approach used to explore the connection between dependent and independent variables. In other words, it serves to analyze how the independent variable influences the dependent variable.

This technique involves analyzing the relationship — even the correlation — between two or more variables in order to identify patterns and trends and capitalize on the projections!

High-Frequency Trading

High-frequency trading (HFT) is a type of algorithmic trading that involves executing transactions at very high speed.

HFT strategies rely on sophisticated algo trading, but also on ultra-high-speed data-transfer networks, in order to execute transactions within fractions of a second. This speed is essential, which

HFT strategies are designed to profit from small price movements in the market. As such, for a trader, using HFT to execute large transaction volumes quickly and efficiently can also cut fees considerably, making it possible to turn a profit in niches that wouldn’t be profitable otherwise.

Indeed, High Frequency Trading, in essence, “means”:

Large transaction volume
A very high number of orders
Low but steady profitability

So it’s essential not to get it wrong when it comes to calculating the break-even point!

I dug up a documentary I watched a while back on high-frequency trading. If this is a topic you’re passionate about, it’s a solid watch for understanding what’s at stake with this strategy!

“In 2 minutes, there can be up to 2 million transactions”

Arbitrage

Arbitrage strategies consist of profiting from price differences between two or more markets.

One of the most common arbitrage strategies is statistical arbitrage. This strategy consists of identifying assets that are mispriced relative to one another, then buying and selling those assets to make a profit.

Statistical arbitrage is designed to capitalize on market inefficiencies and other opportunities.

This term — if I’ve read my audience right — you’ve probably heard more than once, so we won’t dwell on it; let’s just recall a solid definition! Arbitrage was an incredibly lucrative activity as recently as 2 or 3 years ago on the crypto market. In fact, for those who still remember, that’s how SBF started building his fortune through Alameda Research!

Today, arbitrage on the crypto market is probably still profitable, but it has certainly become far more competitive!

In the past, all it took was being registered on a few crypto platforms around the world, holding substantial funds, buying on the platforms where a market imbalance leaned toward supply and selling on the platforms where the imbalance leaned the other way.

Wars, for example, have in recent years created significant arbitrage opportunities on local platforms for anyone — but it’s also a profession, and some people do nothing else!

As a result, in a market whose competitiveness is growing exponentially, AI trading has become indispensable to the practice of arbitrage on the financial markets.

AI Trading and Machine Learning

Predictive Models

Predictive models have almost certainly been used ever since people had information, data, and a method to interpret them. That said, modern predictive modeling reportedly began in the 1940s, when governments used the first computers to analyze weather data, among other things…

Over the following decades, software and hardware capabilities grew, making it possible to store large amounts of data and access it more easily for analysis.

Internet connectivity has made it easier for anyone with access to accumulate, share, and examine vast amounts of data. As a result, modeling has expanded to cover almost every area of business and finance. For example, it’s perfectly common in the marketing world to see predictive modeling used. And of course — this is the part that interests us — financial analysts also use it to estimate trends and events on the stock market.

Uses of Predictive Models

Predictive analytics uses predictors, or identified features, to build models with the goal of reaching an outcome.

Predictive analytics can be used in hundreds, even thousands, of ways. For example, investors use it to spot trends in the stock market or in individual securities that could signal investment opportunities or decision points.

One of the models most commonly used by investors is the moving average of an investment, which smooths out price fluctuations to help them identify trends over a defined period. In addition, autoregression is used to establish a correlation between the past values of an investment or an index and its future values.

Predictive modeling is also a valuable risk-management tool for investors, since it makes it possible to identify the potential outcomes of various scenarios. For example, the data can be adjusted to predict what might happen if a fundamental condition were to change.

The Various Forms of Predictive Modeling

There are several distinct forms of predictive modeling that can be used to analyze most datasets in order to uncover insights about upcoming events.

Classification Predictive Models

Classification models draw on machine learning to sort data into categories or classes according to criteria defined by the user. There are several types of classification algorithms, a few examples of which are below:

Logistic regression: An estimate of whether an event will occur, usually a binary classification such as a yes-or-no answer.

Decision trees: A series of yes/no, if/else, or other binary outcomes presented in a visualization called a decision tree.

Decision tree: An algorithm that combines unrelated decision trees using classification and regression.

Neural networks: Machine-learning models that comb through enormous volumes of data to detect correlations that only emerge after analyzing millions of data points.

Naive Bayes: a modeling system that relies on Bayes’ theorem to establish conditional probability.

Clustering (Grouping) Models

Clustering is a technique for grouping data points together. Cluster analysts therefore assume that data belonging to similar groups should share the same characteristics, and that data belonging to different groups should have very different properties. The most common clustering algorithms are as follows:

K-Means / Mean: K-means is a modeling technique that uses clusters to identify the central tendencies of different groups of data.
Mean-Shift: In the Mean-Shift model, the algorithm shifts the mean of a group so as to identify the “bubbles,” or maxima, of a density function. When the points are plotted on a graph, the data appears to cluster around central points called centroids.
DBSCAN, or Density-based Spatial Clustering With Noise : DBSCAN is an algorithm that groups data points based on an established distance between them. This model builds relationships between the different groups in order to screen out outliers. That said, there are other, more specialized models dedicated to that!

Outlier Models

It’s perfectly normal for a data series to contain outliers. The problem is that these exceptional values can skew our predictions in a fairly unwarranted way. That’s why it’s important to be able to remove outliers.

Isolation Forest: An algorithm that detects the few, deviant data points within a sample.
Minimum Covariance Determinant (MCD): Covariance is the relationship of change between two variables. The MCD measures the mean and covariance of a dataset in a way that minimizes the influence of outliers on the data.
LOF, or Local Outlier Factor : An algorithm that identifies the nearest data points and assigns them scores, which makes it possible to flag the most distant points as outliers.

Time-Series Models

Commonly used before other types of modeling, time-series modeling uses historical data to forecast events.

Here are 2 of the most common time-series models:

ARIMA: The Autoregressive Integrated Moving Average model uses autoregression, integration (differences between observations), and moving averages to forecast trends or outcomes.
Moving Average: The moving average uses the average over a given period — for example the MA50 and the MA200, with 50 and 200 days respectively — in order to smooth out fluctuations.

AI Crypto Trading, Scalping Style!

In the meantime, if you're looking for more in-depth, educational trading content, check out the rest of our free course.

For research purposes, we’ve decided to share this strategy, built from A to Z by a member of the community. The original spreadsheet is no longer distributed, but its logic is broken down below, step by step — everything you need to rebuild and test it yourself. It’s relatively easy to get the hang of, but you absolutely need a bit of practice before going live! And to discuss it with the community, the live chat is a little higher up on this page.

Building a Screener from a Simple Strategy

There are countless ways to build an automated trading system, as well as many levels of automation: you might want to automate everything (from decision-making all the way to order execution) or simply give yourself a personal, complementary tool while continuing to act manually on the market.

Community-Built Automated Strategy

From here, we’ll hand things over to one of our students, who’s going to walk you through an automated model he developed based on USD/USDT disparities to capture trade opportunities on a very short timeframe.

Today, I’m going to introduce you to a screener based on a simple strategy, designed to hand you entries as well as scalp trade setups. The idea is to grab moves on very short timeframes to keep your exposure to an absolute minimum.

You’ve got the full breakdown of my reasoning; don’t hesitate to ping me if you have questions or suggestions for improvements

Test Statistics as of 27/04/24

roughly 1 setup every 3 minutes (across all reliability levels)
On average (the sample is small, so not necessarily representative)
5% strong
16% medium
43% weak
36% kamikaze

Success rate : 64% across all sensitivity levels

Sample: 403 Setups

Context

The USD/USDT Correlation

In theory, 1 USD = 1 USDT; however, temporary divergences can sometimes appear. Several reasons can explain this:

Liquidity
Volatility
External influences (confidence in USDT or USD)

The goal of my strategy is to spot these divergences and profit from them.

Perpetual and Inverse Perpetual Contracts

These are futures contracts with no expiration date, so you can keep a position open indefinitely, as long as you can cover the fees associated with the contract.

The advantage? They let us speculate without having to own the underlying in question.

In our strategy, we use inverse perpetuals as a hedging tool; and finally, one last point worth noting, perpetual contracts also let us use leverage

* Disclaimer: Leverage is dangerous — stay humble, or you risk getting wiped out

Choosing the Platform

For our strategy, we use OKX: the platform offers both USDT-margined perpetuals (the “PERP,” e.g. BTCUSDT) and coin-margined perpetuals (the “Inverse PERP,” e.g. BTCUSD) on the major pairs like BTC, ETH, and XRP

Perfect for our strategy: we can play the USD/USDT correlations on these pairs.

big data algo trading artificial intelligence

Hedging

Simply put, hedging means opening a long position and a short position on the same asset to protect yourself in case the market moves the opposite way from what you anticipated.

Typically: 1 long, 1 short

If the price drops, the short covers the long
If the price rises, the long covers the short

Strategy: Let’s Get Down to Business!

As mentioned earlier, we’re going to build a simple strategy based on USD/USDT divergences, using perp and inverse perp contracts to our advantage, with a dash of leverage on top

The USD/USDT Correlation

1 USD = 1 USDT, most of the time.. the USD/USDT correlation is extremely strong, and the divergence periods, although very short, still exist.

Here, we can see a micro-gap of 8.4 points between BTCUSDT and BTCUSD.

Starting from the premise of the USD/USDT correlation, two conclusions are possible:

USDT could rise to meet USD
USD could fall to meet USDT

So the question is: how do we know which one will meet the other?

Technical Analysis to the Rescue

We know that USD/USDT divergences occur on very short timeframes, so we can use a bit of technical analysis on BTCUSDT to identify a trend.

Note: other technical-analysis methods and other indicators can be used; here, I’m going to give you simple indicators. You can make the technical analysis I’m proposing more complex, modify it, or alter it however you like — the goal is to understand the strategy as a whole.

Correlation of Technical Indicators

For our example, we’ll use the correlation of technical indicators to infer the direction (USDT toward USD or USD toward USDT).

The chosen ones are: MACD / RSI, / Bollinger Bands and EMA

A blend of these technical indicators to define a (very short-term) trend intention on BTCUSDT: this consensus then gives us a sense of the market’s potential direction on this pair. We’re literally playing the probabilities.

The Logic of Our Strategy

If BTCUSDT < BTCUSD
If technical analysis suggests a rise on BTCUSDT
Then BTCUSDT should meet/exceed BTCUSD

Suggested position:

Long Perp BTCUSDT, with leverage to increase the expected gain
Short Inverse Perp BTCUSD, without leverage (for hedging)

Conversely:

If BTCUSDT > BTCUSD
If technical analysis suggests a drop on BTCUSDT
Then BTCUSDT should meet/drop below BTCUSD

Suggested position:

Short Perp BTCUSDT, with leverage to increase the expected gain
Long Inverse Perp BTCUSD, without leverage (for hedging)

Important considerations

Basis risk: The positions won’t fully hedge each other if the USD/USDT ratio doesn’t normalize
Transaction costs: In this kind of strategy, where profits are fragile, you absolutely must factor in the platform’s fees! A trade can look attractive, but once the fees are deducted, it wasn’t worth it
Volatility: Cryptos are extremely volatile, market conditions can be extreme, and liquidity can have a big impact on your trades — better safe than sorry
Your PnL depends on your leverage, your analysis, and your appetite for risk — again, be careful

Finally, an important point to keep in mind: the moves we’re after are on extremely short timeframes, and placing 2 orders correctly within a few seconds is a challenge — ease into it and practice!

The Weakness of This Strategy

The big weakness of this strategy lies in the reactivity it demands. Between analyzing the elements we mentioned above, setting up the 2 trades, and execution, you’ll have your hands full!

For the boldest among you, I’ve put together a cheat sheet that gives you setups based on this strategy. You’ve already seen it go by — it’s right at the top of this article! What you may not have noticed is that it updates automatically !

It’s laid out as follows:

Reliability: Ranging from Kamikaze to High, the reliability tells you about the risk of the proposed setup
Direction : Bullish/bearish
The product: Perp or Inverse Perp
The ticker: BTC/ETH/XRP or EOS
The platform’s Last Traded Price
Order type: Long or Short
TP: Take Profit
SL: Stop Loss

All the elements of the document speak for themselves, but a few of them need some clarification:

The Last Traded Price is the last price traded on the platform for the product (perp or inverse perp); it is updated every minute.

The TP and SL are suggestions; you can fine-tune them however you like. That said, note that here they’re indexed to the ATR and the platform’s fees, and are automatically updated as market conditions evolve

The goal isn’t to give you signals, but rather potential, attractive entries on this kind of setup!

How Do You Place the Orders?

For execution, we use OKX (remember to check availability from your country). The principle stays the same: place your main position (Long or Short) on the USDT-margined perpetual with a bit of leverage, then the opposite order on the coin-margined (inverse) perpetual for the hedge, entering the Take Profit and Stop Loss suggested by the cheat sheet.

Leverage is dangerous: start with a small position size and practice in demo before going live.

What Follows Is Very IMPORTANT for Beginners

For the greenest beginners among you, here are the settings I suggest, along with how to use the sheet while you’re cutting your teeth:

Only trade BTC and ETH
Only place the Long and Short 1 orders, not the hedge order – You won’t have time to place both correctly
Order size at 70 – Without a hedge, you’re not risking much, so nothing too damaging if you mess up
Only take the Medium and High reliability levels
Adjust your trades depending on when you entered – if you see you can move your SL up, do it! Conversely, on the TP, don’t get too greedy

Finally, and more generally, the ideal is to keep the sheet visible on one of your screens, or minimized, so you’re ready to act on BTC and ETH. Pre-fill your order sizes, and let it roll !

You’ll see it works in cycles: quiet > it fires off > quiet > it fires off > etc.

Conclusion

Every trading strategy has a goal and variables; the one I’ve shown you today is meant to support you with scalping and booking small but ultra-regular profits. It won’t suit every profile, but we can say with certainty that it does its job.

If artificial intelligence in the markets and AI trading interest you, don’t hesitate to send me feedback in the comments of this article — I’d be delighted to answer your questions

Finally, a big thank-you to Captain Trading for giving me the opportunity to share my work; we often work alone, and having the team’s support is always a pleasure!

See you on Discord!

FAQ

Frequently asked questions

How do you connect Claude to TradingView?

Through an MCP server (Model Context Protocol), the open protocol Anthropic launched to connect an AI to external tools. The simplest way since September 2026: TradingView's official MCP server. You paste its address into Claude's connectors (or ChatGPT's), sign in to your TradingView account — Essential plan or higher, trials excluded — and you're connected, with nothing to install: the 4-step guide is near the top of this page. The other route, the one behind the 13-minute video tutorial, goes through an open-source MCP server that drives TradingView Desktop: longer to install, but it's the one that lets Claude write Pine Script on your chart.

Can you build a TradingView indicator without knowing how to code?

Yes — that's the whole point of this page. You describe your idea in plain English, Claude writes the Pine Script code, compiles it on TradingView and fixes its own errors. You do need a paid Claude subscription and a paid TradingView plan, though.

Can Claude build and backtest a Pine Script strategy?

It can code a strategy from your rules and run it through TradingView's backtester. Keep in mind that a backtest is only ever a test on the past: it guarantees nothing about real results.

Can the AI trade in my place?

No, and that's not the point. The AI absorbs the repetitive mechanics — the code, the alerts, the watchlists; reading the market, the plan and the decision stay with the trader. That's exactly what the last video on this page shows.

What's next at Captain Trading

We keep testing these tools first, so you don't have to hit the same snags. The official TradingView MCP just came out: its step-by-step guide opens this page and the video tutorials follow it — and the exploration continues live session after live session: new indicators, backtests and custom dashboards.

Watch the videos on this page, then come talk about it on Discord or subscribe to the channel so you don't miss what's next. And if you want to learn to trade with real method, the free trading course is a solid place to start.

Already happening in the community

So what's next? Step through the door of grid bots

A grid bot is a robot that places buy and sell orders at regular intervals inside a range that YOU define ahead of time. The repetitive mechanics, taken to their purest form — and still the same philosophy as the rest of this page: the tool executes the plan, the trader sets the frame, the levels and the risk. It's already being discussed in the community's dedicated channel: questions, real feedback and settings get shared there between members.

Join the Discord →

Browse the free trading course → The YouTube channel

Content published for educational and informational purposes only. It does not constitute investment advice in any way. Trading carries a risk of losing capital.

Claude × TradingViewthe AI that codes your Pine Script indicators(and what it will never do)

The essentials

Official TradingView MCP: it's finally here!

The tutorials — connecting Claude, then putting it to work

AI trading: what are we talking about?

1. Rule-based bots (automated trading)

2. Quantitative models and machine learning

3. Generative AI as copilot (LLM)

4. Regulated robo-advisors

AI trading tools in 2026: the comparison

What we actually plugged in

Pine Script without knowing how to code: what it lets you do

The replay — the indicator coded live

The limits, no filter

How much AI trading costs, line by line

1. Testing it out, like in the tutorials and the guide

2. Running it every day

3. A rule-based bot on an exchange

What the AI won't do in your place

The latest AI videos

AI Trading: the complete guide

AI Trading: the Guide!

AI Trading: Definition

AI Trading: the Main Uses and Technologies

AI Trading: the Advantages

Algo Trading: the Different Types of Algorithmic Trading

Quantitative Analysis

High-Frequency Trading

Arbitrage

AI Trading and Machine Learning

Predictive Models

Uses of Predictive Models

The Various Forms of Predictive Modeling

Classification Predictive Models

Clustering (Grouping) Models

Outlier Models

Time-Series Models

AI Crypto Trading, Scalping Style!

Test Statistics as of 27/04/24

Context

The USD/USDT Correlation

Perpetual and Inverse Perpetual Contracts

Choosing the Platform

Hedging

Strategy: Let’s Get Down to Business!

The USD/USDT Correlation

Technical Analysis to the Rescue

Correlation of Technical Indicators

The Logic of Our Strategy

The Weakness of This Strategy

How Do You Place the Orders?

What Follows Is Very IMPORTANT for Beginners

Conclusion

Frequently asked questions

How do you connect Claude to TradingView?

Can you build a TradingView indicator without knowing how to code?

Can Claude build and backtest a Pine Script strategy?

Can the AI trade in my place?

What's next at Captain Trading

Claude × TradingView
the AI that codes your Pine Script indicators(and what it will never do)