Article
How to Actually Use AI to Trade Better, a Practical Claude Code & Codex Workflow
Stop pasting screenshots and vague prompts into AI. Here is the exact workflow for using Claude Code or Codex with your real chart data to review your process, build journals, and draft indicators, one reviewable job at a time, never trade decisions.
Most traders 'use AI' like this: screenshot the chart, paste it into Claude or ChatGPT, ask 'is this a good trade?', and get a confident, generic answer they cannot check. The tool is not the problem. The handoff is. A screenshot is just pixels, and a vague prompt invites a vague guess. This is the exact workflow that turns AI from a party trick into something that measurably sharpens your process, using your real chart data, one reviewable job at a time.
Why 'here is a screenshot, is this a good trade?' fails
A screenshot is pixels. The model cannot read your exact levels, does not know your timeframe, your rules, or your risk, so it infers all of it and answers confidently anyway. You get a polished paragraph built on guesses.
The same trap hides in 'build me a trading bot.' The output looks finished and is impossible to verify. You do not need a smarter model. You need a better handoff.
The one idea that changes everything: give AI an input contract
Every useful AI trading task has the same three parts: structured data it can actually read, one narrow job, and acceptance checks so you can challenge the output. Miss any one and you are back to confident guesses.
| Weak handoff | Strong handoff | What you get back |
|---|---|---|
| A screenshot | A structured chart export (levels, OHLC, notes) | The 6 levels you drew and which ones price tested |
| 'Is this a good trade?' | 'Did my entries follow my own rules?' | A flagged list of rule violations |
| 'Build my system' | 'Draft a parser for these journal fields' | Reviewable code plus its stated assumptions |
| A vague indicator idea | 'Shade the opening range, alert on a break' | A custom formula indicator, built |
Step 1, Turn your chart into data the AI can read
MyLinedChart exports your drawings, levels, notes, OHLC, volume, indicators, and session fills as JSON, XLSX, or CSV. Now the AI sees resistance 750.95 and opening-range low 749.15 as real fields instead of guessing at pixels.
Or skip the copy-paste entirely: connect your AI directly at MCP toolkit. It reads your live chart context and can even draw on it, with every change confirmed by you before it lands.
Try it: 'Here is my SPY session export (JSON). List every level I drew, its price, and which ones price actually tested today.'
Step 2, Give it a narrow, reviewable job
Scoping is the skill. One job at a time. Each of these is something AI genuinely does well when you hand it structured data:
- Review your process: 'Compare my planned levels to where I actually entered, flag where I chased.'
- Build a journal: 'Turn these exports into a table: date, symbol, setup, planned entry, actual entry, R multiple.'
- Draft a custom indicator: 'Write a formula indicator that shades the opening range and alerts on a break.' MyLinedChart builds it from that sentence.
- Draw today's levels: 'Fetch my daily-levels JSON and draw them on my chart.' You approve the draw.
- Audit rule discipline: 'Here are my 20 trades and my rules. Which trades broke a rule?'
Step 3, Write prompts that force the AI to show its work
Add one line to any prompt: 'List your assumptions, any fields you are missing, and how you would test this answer.'
That single instruction is the difference between confident garbage and reviewable help. It turns the AI from an oracle into an assistant that hands you its reasoning to check.
A worked example: from a messy session to a review you trust
Export your SPY session, paste the JSON to Claude Code, and ask for a planned-versus-actual table plus rule-violation flags.
Claude returns the table and adds: 'Assumption, I counted any entry more than 2 ticks from your drawn level as a chase. Confirm?' You confirm. It shows you chased 3 of 5 entries.
That is the whole loop: your data in, one narrow question, a checkable answer, a lesson for tomorrow. No signal-guessing anywhere in it, just a faster, more honest review of what you already did.
What to never ask AI to do
Do not ask it to decide what or when to trade, size a position, approve an order, or invent data it does not have.
MyLinedChart's AI connection is read-and-annotate, confirmation-gated, and never places orders, account and order data stay read-only. The human stays in the loop by design, not by hope.
Where this leaves you
Structured exports, one narrow job, and review checks add up to AI that actually improves your process instead of impersonating a guru. Start with a single export and a single question.
For the IBKR-specific setup, see Using Codex or Claude Code With IBKR Chart Data. To pick between the two tools, see Codex vs Claude Code for IBKR Chart Data Workflows. Connect your AI at MCP toolkit, or start a trial at Pricing. Want it wired to your exact workflow? See workflow consulting, planning support only, no trading advice.
FAQ
Can AI tell me if a trade is a good idea?
No, and you should not want it to. It cannot see the future or your full context. But it can check whether a trade you already took followed your own written rules, which is far more useful for improving.
Do I need to be technical to do this?
No. You describe what you want in plain language; MyLinedChart produces the structured export, and the AI does the parsing, tables, or indicator build. You review the result.
Claude Code or ChatGPT Codex, which should I use?
Either works with structured chart exports. See Codex vs Claude Code for IBKR Chart Data Workflows for the trade-offs. Both are third-party tools with their own subscriptions.
Is my data safe?
Your exports are yours, and the connection runs locally against your own broker session. The AI only sees the chart context you hand it, and it can never see or touch your orders.
What is the first thing I should try?
Export one trading session and ask the AI for a planned-versus-actual comparison. It is the fastest way to see the workflow pay off.
Sample Structured Chart-Data Exports
Review how chart drawings, annotations, OHLC, volume, and execution context become reusable structured data.