You’ve seen people talk about Claude for trading, and you want to try it on your own chart without writing a line of code. A Claude Opus trading bot in MT5 needs two things: a model that can read the market context, and an EA that can send it that context and execute the decision. Here’s how I set it up, step by step, in about ten minutes once you have your API key.
I’ll use Alpha Pulse AI, the EA I build, because that’s the one I know. The steps that involve Anthropic and MetaTrader are the same for any EA that calls Claude.
What you need before you start
- MetaTrader 5 (or 4) on a demo account. Demo first, always.
- An Anthropic account with prepaid credits. Anthropic works with a credit balance, not a monthly invoice.
- Alpha Pulse AI v3.03 or later, which includes the Claude Opus 5.5 preset.
- Ideally a VPS. Most users run the EA on one, because the EA has to be online when the market moves.
Step 1: get your Claude API key
- Sign up at console.anthropic.com.
- Add prepaid credits under Billing.
- Create a key in Account Settings, then API Keys, and copy it somewhere safe.
The key belongs to the provider of the model you select. A Claude model needs a Claude key; an OpenAI key won’t work with it.
Step 2: allow WebRequest in MetaTrader
In MetaTrader, open Tools, Options, Expert Advisors, and tick “Allow WebRequest for listed URLs”. Add https://api.anthropic.com to the list. Without this, the EA can’t reach Claude and nothing happens.
Step 3: choose Claude Opus 5.5 in the EA
The inputs are laid out as a wizard in eight numbered sections. For this setup you only need the first four:
- Preset: the strategy engine. Institutional Auction and Liquidity is the default recommendation for a single symbol.
- AI Model and API Keys: paste your key in the “Claude API Key” field and select Claude Opus 5.5 in AI Model Selection. The other Claude options are Fable 5.1, Sonnet 5 and Haiku 4.5.
- Risk Per Trade: the default is 1% of the balance. On a demo, keep it there or lower.
- AI Behavior and Cost: set your Reasoning Effort and your daily API budget.
Anthropic released Claude Opus 5.5 on 22 September 2026. Its thinking is always on: the Reasoning Effort input sets how deep it thinks, and Disabled runs it at its default depth. I tested it live on my setup with Disabled and with Medium before adding it to the EA.
Step 4: put a ceiling on the API bill
This is the step people skip. Every decision is an API call, and a strong model on a busy chart adds up. Max Daily API Cost is a hard ceiling per chart and model: when the day’s budget is spent, the EA stops asking. It never silently switches to a cheaper model behind your back. Claude requests also use prompt caching, which keeps repeated context cheaper.
Start with a small daily budget on demo and look at what a normal day really costs before you raise it.
Step 5: forward test, because you can’t backtest it
The Strategy Tester can’t run this. The EA talks to Claude through WebRequest, and WebRequest isn’t executed in the tester. So the only honest test is forward: on a demo account, for long enough to see different market conditions.
No Alpha Pulse yet? Its Market demo hits the same wall, because it only runs in the Strategy Tester. So there’s a free 7-day trial for your own demo account: follow this exact guide with it, Claude Opus 5.5 included.
Record the model, the prompt or preset, the symbol, the timeframe and the risk settings before you start, and don’t change them mid-test. Here’s what to record in an AI trading forward test.
What a Claude Opus trading bot does, and what it doesn’t
| It does | It doesn’t |
|---|---|
| Read the chart context the EA sends and return a decision with its reasoning | Guarantee a profitable strategy |
| Follow your instructions and the EA’s risk settings | Replace your risk management |
| Wait when the setup isn’t clear (if your rules say so) | Work inside the Strategy Tester |
| Change model when a better one comes out, without changing your process | Cost nothing: every decision is a paid API call |
In my own live comparisons, Claude models have been strong at reasoning through ambiguous setups. Gemini has been faster and steadier under load. I wrote about that trade-off in Opus 4.7 vs Gemini 3.1 on a live account. Test Opus 5.5 against your current model on demo with the same prompt and risk; that comparison is worth more than any benchmark.
See a new model go into the EA: in Jev AI in MT5: I Added It. Here’s What It Does, I show what the model decides and what the EA keeps under its own control, and at the end I mention that Claude Opus 5.5 is in Alpha Pulse AI too.
Watch out for fake “Claude bots”: in GPT-5.5 Is Here. Fake AI Trading Bots Are Coming Too, I show how to spot an “AI bot” that never calls the model at all. Every big release brings a wave of them.
My ten-minute checklist
- Anthropic key created, credits added.
- WebRequest allowed for api.anthropic.com.
- Claude API Key pasted, Claude Opus 5.5 selected.
- Risk per trade at 1% or less, on demo.
- Reasoning Effort chosen, Max Daily API Cost set.
- Configuration written down, forward test started.
Before buying any AI EA, mine included, run through what to check before you buy Alpha Pulse AI. And if you’re still choosing a provider, the best AI for MT5 EAs compares them.
If your longer-term plan includes allocated capital, keep that decision separate from the model. My Axi Select overview explains that route. Permitted automation needs its own check there. Affiliate link; I may earn a commission if you join through it.
Programmes like that deserve the same patience as a new model. Read the requirements and costs next to the opportunity; the Axi Select breakdown helps you do that.
I test every new model on demo before it goes near a live account, and I share what I find. Join the DoItTrading newsletter to get it.
Trading with AI EAs involves real risk to capital. API costs depend on your settings and usage. Results of any model or configuration don’t guarantee future ones.
Frequently Asked Questions
Can Claude trade on MT5?
Not on its own. Claude needs an EA that sends it the market context and executes its decisions. Alpha Pulse AI does that with your own Anthropic API key and the Claude model you select.
Which Claude model is best for trading?
Alpha Pulse AI offers Fable 5.1, Opus 5.5, Sonnet 5 and Haiku 4.5. Stronger models reason more but cost more per decision. Run your candidate against your current model on demo with the same prompt and risk.
How much does a Claude trading bot cost to run?
Every decision is a paid API call, so it depends on the model, how often the EA asks and how much context it sends. Set Max Daily API Cost, run a demo week and read the real bill before you raise it.
Do I need a VPS for a Claude trading bot?
Most users run it on one, because the EA has to stay online while the market moves. A home computer that sleeps or loses connection will miss decisions.
Can I backtest a Claude trading bot in MT5?
No. The connection to Claude uses WebRequest, which isn’t executed in the Strategy Tester. Forward test on a demo account instead.
Can I try Alpha Pulse AI before buying?
Yes. The Market demo can’t run AI, because it only works in the Strategy Tester. There’s a free 7-day trial for your own demo account, with your own API key.
