Build Your Own Automated Options Trading System · Bonus
Test a strategy by talking
about it.
Volrix is a backtesting engine. Claude Code can drive it for you. You describe a strategy the way you'd explain it to a friend — and a few minutes later you have real numbers from seven years of market data.
First — what is this, really?
You already know the hard part.
You know what you want to test.
Backtesting used to mean: learn Python, buy data, clean the data, write the strategy, fix the bugs, run it, read the output. Five skills before your first answer.
Now it means: type what you want in English. Volrix already has the data. Claude already knows how to write the code. You supply the idea and the judgement.
You do not need to understand how it works.
You will never see the code, the data files or the plumbing — unless you ask to. This lesson only teaches you what to say, and how to tell a good answer from a bad one. That is the whole job.
Step one · connect it
Four clicks. You do this once, ever.
1
Have a Volrix account
Sign up on the Volrix website and log in. Your plan decides how many backtests you can run — check that first so you're not surprised later.
2
Add Volrix as a connector on claude.ai
Open claude.ai → Settings → Connectors → Add custom connector. Paste the Volrix address: https://mcp.volrix.ai/mcp
3
Log in when it asks
A Volrix login window pops up. Approve it. That's the handshake — it tells Volrix that this Claude is allowed to use your account.
4
Switch it on inside Claude Code
In your terminal, type /mcp. You'll see a list of connectors. Make sure Volrix says Connected. If it says disabled, turn it on there.
That's the entire setup. Nothing to install, nothing to configure, no keys to copy.
Step two · check it's alive
Before anything else, ask it one easy question.
Type this
"How far back does your NIFTY data go?"
NIFTY data is available from 11 February 2019 to 21 August 2026.
That's 1,860 trading days.
If you get an answer like this, everything works. If you get an error or Claude says it can't reach Volrix, fix that now — nothing else will work until this does.
Never skip this. Ten seconds here saves twenty minutes of confusion later.
Step three · what you can test
Every strategy is one of four shapes.
⏰
Same day
You get in and out on the same day. Nothing is held overnight.
"Sell the straddle at 9:20, close everything at 3:10."
🌙
One night
You enter near the close and exit the next morning. One sleep, that's it.
"Buy at 3:15, sell at next day's open."
📅
Many days
You hold for days or weeks — till expiry, till a target, till a stop.
"Sell the monthly strangle 15 days before expiry, hold to expiry."
👀
Just look
No trades at all. You only want to know how often something happens.
"How often does NIFTY move 1% after 2pm?"
The fourth one is the most underrated.
Before you test a strategy, you can just ask the market a question and get a table back. Most bad strategies die at this stage — cheaply, before you've built anything.
Step three · what you can test
If you can name it, you can probably test it.
Stop losses, targets, trailing stops and re-entry after a stop are all standard. Just say what you want.
Step three · the honest limits
Five limits on what it can trade.
One strategy trades one index. A run is locked to one index's option chain, expiry and lot size, so there's no NIFTY-vs-SENSEX pairs trade. Testing both is normal though — ask for both, get two runs side by side. And a strategy can freely watch any other index for signals: reading is unrestricted, only trading is locked.
The smallest slice of time is one minute. There is no tick-by-tick. If your idea depends on what happened in a 3-second window, this cannot answer it — and honestly, neither can your broker on a bad day.
Volume indicators don't work on the index itself. NIFTY and SENSEX spot candles carry no volume figure — and neither does INDIA VIX. So VWAP, and anything else built on volume, has nothing to read there. If your idea needs volume, point it at futures, options, or an ETF like NIFTYBEES or BANKBEES instead.
The last two minutes of the day aren't tradable. The engine's day runs 09:16 to 15:28, and its last candle reads the 15:27 price — that's what "at the close" gets you, for entries and exits alike. The market's own 15:28 and 15:29 prints are in the data, but the engine won't transact on them. Both sides use the same convention, so comparisons stay fair — just don't build a strategy that needs the closing seconds.
BANKNIFTY has monthly expiry only in the index segment. Weekly BANKNIFTY plans need rethinking as monthly.
Read anything. Trade one thing. That's the shape of it.
Step three · the honest limits
And four facts about the tool itself.
The index freezes at 3:15pm; the close lands at 3:29. From 15:15 the cash market is in its closing auction, so NIFTY and SENSEX sit at one frozen value for fourteen minutes — that flat line is not real price action. The official closing price then appears in the 15:29 bar. Options and futures carry on trading normally right through to 15:40; it is only the index that stops.
Data starts in Feb 2019. You cannot test 2008, or the 2020 crash entry. Seven and a half years is what you get — about 1,860 trading days.
Results vanish after about two hours. The detailed trade data isn't kept forever. Download the CSV the same day, or you'll have to run the whole backtest again.
It is not a trading system. It never places an order. It cannot see your broker or your money. It is a very fast history book, nothing more.
A tool that tells you its limits up front is a tool you can trust with the rest.
Step three · the costs it leaves to you
Four things that quietly flatter your results.
Slippage starts at zero. The test assumes you got filled at the exact price on the screen, every single time. You have to say what slippage to assume — and it's one number applied to both your entry and your exit. You cannot set 0.5% going in and 1% coming out.
Brokerage, STT and charges are off. Unless you ask for them to be included, every figure you see is before costs. On a high-frequency strategy this alone can turn a winner into a loser.
Capital is assumed to be Rs 1,50,000. The tool's standing default — almost certainly not your number. Tell it your real capital, or every percentage return is against the wrong base. It never checks margin either: capital is applied when reading results, long after the trades were placed.
Stop losses are assumed to fill at exactly your price. Your SL is checked against every one-minute candle's high and low — so a spike inside the minute does trigger it, which is good. But the exit is then booked at the SL level itself. On a fast move your broker gives you a worse price than that. Every stop-loss exit you see is a best case.
The good news: you don't have to re-run anything.
The first three are applied after the run. Say "show me that again with 1% slippage and brokerage included" and you get a fresh answer in seconds — without spending another backtest.
Step four · the important skill
A good prompt answers nine questions. That's all.
No special words needed — just leave nothing to guess. Anything you don't say, Claude has to assume.
1. What? NIFTY, SENSEX or BANKNIFTY. Pick one.
2. When do I enter? A clock time, or a condition.
3. Buy or sell what? CE, PE, both. Which strike.
4. How much? Number of lots. Always lots, never quantity.
5. When do I exit? A clock time, or a rule.
6. What's my stop loss? Percentage, or a price. Or say "none".
7. Which days? Every day? Only expiry day? Only Mondays?
8. Over what period? "Last two years", or exact dates.
9. What slippage and charges? Say a number — e.g. "1% slippage, include brokerage". Otherwise both are zero.
One more line, and it's the most valuable one you'll ever type:
"Ask me about anything I haven't specified, instead of assuming." — turns a guessing machine into a checklist. Use it every time.
Step four · see the difference
Same idea. Two prompts. Very different results.
Weak prompt
"Backtest a short straddle on NIFTY for me."
- Which strike? It'll pick ATM. Maybe you wanted OTM.
- What time? It'll pick 9:20. Maybe you wanted 11:00.
- Stop loss? It might use none. Your real account has one.
- Which days? It'll test every day. You may have meant expiry day only.
- How long? Unclear — and the period decides the answer.
- You get a number. You don't know what it's a number of.
Strong prompt
"Backtest a NIFTY short straddle. Sell ATM CE and PE at 9:20am, 1 lot each. 30% stop loss on each leg. Square off at 3:10pm. Only on expiry days. Last three years. Ask me about anything I haven't specified instead of assuming."
- Every knob is set by you.
- The result is reproducible — run it again next month, same answer.
- You can change exactly one thing and know what caused the difference.
It took you eleven extra seconds to type. It's the difference between a result and a rumour.
Step five · the real power
The first test is never the point.
The next twenty are.
Once one backtest exists, changing it is one sentence. Say these:
Change one thing
"Same thing, but enter at 9:45 instead."
"Same thing, but with a 50% stop instead of 30%."
"Same thing, on SENSEX."
Sweep a range — ask for the gaps
"Now run it at 9:15, 9:45, 10:15, 10:45 and 11:15, and show me all five side by side in a table."
The golden rule: change one thing at a time.
If you change the entry time and the stop loss together and the result improves, you have learned nothing — you don't know which one did it. Slow is smooth, smooth is fast.
Step six · reading the result
Six numbers come back. Here's what they actually mean.
Net P&L
What you'd have made over the whole period, after costs. For the number of lots you specified — not per lot unless you asked.
Max drawdown
The worst losing stretch, peak to bottom. This is the number that decides your capital. If you can't survive it, the profit is imaginary.
Win rate
How often it won. Beware: 90% win rate with a huge drawdown means picking up coins in front of a bus.
Calmar
Profit divided by worst drop — reward per unit of pain. Above 2 is respectable. This is the best single number for comparing two strategies.
Sharpe
How smooth the ride was. Higher means fewer sickening swings for the same return.
Number of trades
Your sample size. Under about 100 trades, be suspicious. Twelve great trades is luck, not evidence.
Look at drawdown before profit. Everyone does it the other way round, and everyone regrets it.
Step seven · getting the CSV
Want it in Excel? Just ask for it.
Type this
"Give me a CSV download link for all the trades from that backtest."
↓
You get back a link. Click it — the file downloads.
Open it in Excel or Google Sheets. Every trade: date, time, which option, entry price, exit price, profit or loss.
3
Three different files you can ask for
The trade list — every entry and exit. · The day-by-day P&L — for plotting your own equity curve. · The analysis output — if you ran a "just look" study.
Two things to remember about the link
It expires in 30 minutes — if it's dead tomorrow, that's normal, just ask for a fresh one. And anyone with the link can open it without logging in, so don't paste it into a group chat. Always ask for the download rather than having the trades printed into the chat.
Step eight · things that will bite you
Fast answers are easy. Correct answers need these four habits.
If it looks amazing, get suspicious — don't get excited. Test the settings on either side of your winner. A real edge still works at 9:15 and 10:15, not only at 9:45. If it works at exactly one setting, you found a coincidence.
Never take the best number as your expectation. You picked the best of five — of course it looks good. Plan on the average of the winner and its neighbours. That's the honest number.
A backtest is not a settings file. It tells you an idea is worth building. It does not tell you what to type into your live system. That decision is yours, made later, on purpose.
Backtests assume perfect fills. No rejections, no queue, no broker outage, no widened spread on a bad morning. Real life will take a slice off every number you see.
The tool got twenty times faster. Your judgement did not. That gap is where the money is lost.
Do this today
Your first hour, in order.
1
Connect Volrix and ask how far back the data goes.
Ten seconds. Proves everything works.
2
Test the simplest strategy you already trade.
Not a new idea — something you already do, so you can sanity-check the answer against your own experience.
3
Change the entry time. Only the entry time.
Now you've learned the loop that matters.
4
Ask for all five times in one table.
This is the moment most people realise what they've been missing.
5
Download the CSV and look at the ten worst trades.
The losers teach you more than the total ever will.
Takeaway
You don't need to know
how the machine works.
You need to know
what to ask it.
Say exactly what you mean. Change one thing at a time. Distrust the best number. Look at the drawdown first. Everything else is just typing.
End of Backtest by Talking
Next: from a good result to one lot.
A result you believe → a written plan → paper trade → one lot →
Want the detailed technical version of this lesson? /tc/concepts/volrix-mcp