Axi Clients Can Now Ask Claude to Place Trades in Their MetaTrader Accounts

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Axi said this week that eligible clients can connect their MetaTrader 5 (MT5) accounts to Claude, Anthropic's AI assistant, and use it to place trades. The link runs through the Model Context Protocol (MCP), the open standard that lets AI tools talk to outside data and services. Finance Magnates broke the story, and on the surface it lands in the AI trading bot sub-niche we cover every quarter. The qualifier matters, though, and the marketing tends to blur it. This is a natural-language execution layer, not a strategy engine. Axi's own help center says Claude "is not an automated strategy or Expert Advisor (EA)."

We have benchmarked Axi's connector against Zephyr AI's adaptive engine in our 2026 review cycle, and the first thing our process flags is not the AI at all. It is the permission model sitting underneath it, because that is where the risk actually lives.

What does Axi's Claude connector actually do?

Stripped of the AI framing, the connector does a fixed set of things. Per the connector's website, it can check live prices, open, modify, and close positions, place pending orders, and adjust stop-loss and take-profit levels. That is a standard MT5 order-management surface, exposed to a chat window. Setup is manual. You add Axi as a custom connector in Claude, enter the broker's MCP address, and approve access by logging in with your Axi credentials. Axi support can guide you through it but, per the help center, cannot set the connection up on your behalf.

The eligibility rules are narrower than the headline suggests, and that matters for anyone running multi-account setups. Only active MT5 accounts qualify, and they can be verified or unverified. Accounts that run only on MT4 or on the Axi trading platform are excluded. In our 2026 algorithmic testing framework, we treat platform coverage as a hard gate, because a connector that only speaks MT5 cannot be evaluated on the same footing as a cross-platform execution layer. If your book is split across MT4 and MT5, half of it is invisible to Claude.

Axi describes the service as being in alpha testing. That is the most important word in the announcement. Alpha means the order-routing path is not yet a production system, and we would want the connector's error handling and reconnect logic documented before trusting it with anything beyond a demo account.

Is this a trading bot or an assistant?

This is where the AI trading bot label does real damage. Claude is a large language model. It is not a rule-based expert advisor with published entry and exit logic, a fixed lot-sizing rule, and a reproducible backtest. Axi's help center is explicit that Claude is not an automated strategy or an EA, and that it acts only when a client asks, unless the client has set a specific action to "Always allow" in Claude's permission settings.

We separate two things the industry keeps merging. Machine learning, in the trading sense, means a model trained on market data to output a signal or a size. A rule-based system means explicit if-then logic. An LLM chat connector is neither, strictly speaking. It is an intent interpreter sitting on top of an order API. When a vendor calls that "AI-powered trading," we ask for the training set, the feature list, and the out-of-sample window. Here there is nothing of the sort to ask for, because the product is not making the trading decision. You are. That is honest, and Axi deserves credit for saying it in its own documentation rather than letting the headline carry the claim.

The "Always allow" setting is the crack in that honesty. It converts a per-action approval tool into a standing permission. We flag it because it is the one configuration where the assistant stops behaving like an assistant and starts behaving like an unattended executor.

How does Axi compare with other brokers' AI connectors?

Axi is not first to this. The MCP land grab started when MetaQuotes added native AI agents and MCP support to MT5 in a July beta, as Finance Magnates reported. Since then, at least six other brokers or vendors have shipped something comparable, and the differences are instructive.

The pattern is clear. Axi and Leverate both put a human approval step in front of every order. Interactive Brokers lets Claude draft trades but requires manual sign-off on each order, a setup it launched in June for equities and ETFs, per Finance Magnates. ThinkMarkets' ChelseaAI server lets AI models execute orders on ThinkTrader but not access funds, the broker said at launch. Match-Trade opened its MCP to retail traders on October 5, with an option for compatible clients to require approval for each action, per Finance Magnates. CMC Markets went the other way entirely. Its ChatGPT link for UK clients is read-only and cannot trade, change orders, or move money, Finance Magnates reported.

So the market has split into three camps. Read-only assistants (CMC), draft-and-sign (IBKR), and execute-with-approval (Axi, Leverate, ThinkMarkets, Match-Trade). Axi's real differentiator is cost, not capability. Which brings us to the fee question.

What does it cost, and how does that shape strategy economics?

The entry cost on the AI side is zero. Axi says the connector works on Claude's free plan, which allows one custom connector. TradeStation, by contrast, requires a Claude Pro subscription and a $10,000 account balance for its MCP link, according to the source report. That is the sharpest cost contrast in the current field.

Free Download: Claude + Axi MetaTrader AI Trade Execution Due-Diligence Checklist
A step-by-step checklist to vet Claude's MetaTrader trade placement on Axi, covering permissions, broker compatibility, fees, regulatory status, and withdrawal flow before you connect real capital.
Get the Axi Claude Checklist

For a $5,000 retail account, the TradeStation gate is disqualifying on its face, since the balance requirement is double the account. Axi's zero-cost entry removes that friction. But free access is not the same as cheap trading. The connector does not change your spread, your commission, or your swap. It changes who is typing the order. If your strategy is spread-sensitive, the AI layer is economically neutral, and the real cost question remains the broker's execution pricing, which we would verify directly against the Axi spread schedule rather than infer from the announcement.

Not sure which AI trading bot fits your strategy? Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership - see our editorial policy for details.

Where does the risk actually sit?

There is no published backtest here, and we will not pretend otherwise. A chat-executed order has no historical equity curve, no Sharpe ratio, and no maximum drawdown to quote, because the human, not the model, sets the size and the timing. In our testing framework we would normally re-implement a strategy and run walk-forward across a multi-year window. That is impossible with a natural-language connector, and any review that quotes a drawdown figure for it is inventing one. Verify with the provider if you see a number attached to this product.

The risks that do exist are operational. First, approval latency. Every order waits for a human "yes" in chat, which is fine for swing entries and fatal for anything that needs sub-second fills. Second, the "Always allow" override, which we would disable by default. Third, the alpha status, which means reconnect behavior after a dropped MCP session is not yet battle-tested at scale. Fourth, and least discussed, the model's non-determinism. The same prompt can produce different order phrasing on different days, so the audit trail is conversational rather than structured.

In our 2026 testing program we run every execution layer through one test: can it produce a structured, timestamped order log that reconciles to the broker statement line for line? The Axi connector, as documented, produces a chat transcript, not a structured log. That is a gap we would want closed before alpha ends, and it is a gap a rule-based engine does not have, because its decisions are deterministic by construction.

Can you disconnect cleanly?

Disengagement is where chat connectors can quietly outperform traditional EAs. To stop an EA you have to remove it from the chart, confirm it is not running on another symbol, and check that no pending orders are orphaned. To stop the Axi connector you revoke the custom connector in Claude and, if you had enabled it, clear the "Always allow" permission. That is a two-step teardown, and it is cleaner than most EA removals we have logged.

The caveat is pending orders. If Claude placed a limit order and you then revoke the connector, the order lives on in MT5 until it fills or you cancel it. Revoking access does not cancel working orders. That is a real footgun, and it is the kind of thing that should be spelled out in the help center rather than left to the user to discover after a fill.

Is Axi regulated, and does that matter here?

Axi operates in multiple jurisdictions, and the connector does not change that. We do not assert a license number we cannot cite, so treat any specific claim as something to verify directly with the provider's primary regulator. The relevant registers are the FCA Register for UK permissions and the ASIC Connect register for Australian authorisations. If you are placing trades through a chat window, the entity that holds your account and the entity that built the connector are the same, and that entity's regulatory status is the one that governs your dispute path.

The regulatory edge case worth watching is not the broker. It is the AI provider. Anthropic is not a licensed financial adviser, and Axi's help center says Claude is not a financial advisor and that its answers are not trade recommendations. So when a client asks Claude whether to buy EURUSD, the answer is not a regulated recommendation, and no one is on the hook for it. That is a genuine gap in the framework, and it will get tested the first time a retail client loses money on a chat-suggested trade and files a complaint.

Why the Always allow setting is the real risk

Here is the under-discussed risk we think the source material missed. The industry is debating whether AI should be allowed to trade. The more immediate problem is permission drift, where a tool sold as per-action approval quietly becomes an unsupervised executor through a settings toggle. Axi's "Always allow" is one example. Once a client enables it for, say, closing positions, the connector can close positions without asking. Stack that with a looping prompt and you have an EA in everything but name, running on an alpha-stage order path, with no backtest, no drawdown limit, and no kill switch beyond revoking the connector.

We have seen this movie before with copy trading. The platform ships a safe default, users disable it for convenience, and the risk profile changes without anyone updating the disclosure. The fix is boring and effective: default "Always allow" to off, require a typed confirmation to enable it, and surface a persistent banner in the chat when it is on. None of that is technically hard. It is a product decision, and the brokers that make it first will own the trust narrative. Compare that with Zephyr AI's adaptive engine, where position sizing and risk limits are enforced by the engine rather than left to the user's chat discipline.

How Zephyr AI compares

Where Axi's connector is an intent layer with no strategy underneath, Zephyr AI's adaptive engine is the opposite design. Position sizing and risk limits are baked into the engine rather than left to the user's typing. On the dimension that matters most to us, drawdown control, an engine that enforces its own limits is structurally safer than a connector that will execute whatever you type, including a size you fat-fingered at 2 a.m. That is not a knock on Axi's execution. It is a statement about category.

If you want to place discretionary trades by talking, Axi's zero-cost, approval-first connector is a reasonable alpha-stage tool, and its free-plan entry beats TradeStation's $10,000 gate outright. If you want a rules-based system with a defined risk envelope, you need an engine, not an assistant, and that is where Zephyr AI's adaptive position sizing is the cleaner fit.

Not sure which AI trading bot fits your strategy? Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership - see our editorial policy for details.

Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Does the Axi and Claude connector work in the United States?

Axi's announcement does not spell out US eligibility, and the connector's own documentation limits access to active MT5 accounts. If you trade a US-regulated account, verify directly with Axi whether your entity is in scope before wiring anything up. We would not assume availability from a global press release.

Can I run it on a prop firm account?

The connector requires an active Axi MT5 account, so a prop firm running its own MT5 infrastructure is a separate question. Even where it is technically possible, most prop evaluations prohibit third-party order automation, and "Always allow" would almost certainly breach those terms. Check the prop firm's rulebook first.

Is Claude actually placing the trades, or just drafting them?

It executes, but only after you approve. Axi says any trade or order action needs the client's explicit approval in the chat before it goes through. Interactive Brokers, by contrast, lets Claude draft trades but requires manual sign-off on each order. The practical difference is small; the disclosure difference is not.

What happens if the MCP connection drops mid-trade?

Axi describes the service as alpha, and the documentation does not publish a reconnect specification. A dropped session should not close an open position, since the position lives on the MT5 server, but any instruction in flight would be lost. Test the reconnect path on a demo account before trusting it live.

Do I need a paid Claude subscription?

No, per Axi. The connector works on Claude's free plan, which allows one custom connector. TradeStation's MCP link, by comparison, requires a Claude Pro subscription and a $10,000 account balance.

Can Claude access or withdraw my funds?

Axi's connector covers prices, positions, pending orders, and stop and take-profit levels. It does not move money. ThinkMarkets makes the same separation explicit, letting AI execute orders but not access funds.

Which accounts can connect to the Axi connector?

Only active MT5 accounts, verified or unverified. Accounts that run only on MT4 or on the Axi trading platform are excluded, so a multi-platform book is only partly visible to Claude.

Is the Axi connector a trading bot or an expert advisor?

Neither, by Axi's own description. Its help center states Claude is not an automated strategy or EA and acts only when asked, unless a specific action is set to "Always allow." Treat it as an execution assistant, not a strategy.

How do I turn the connector off?

Revoke the custom connector in Claude and clear any "Always allow" permissions. Remember that revoking access does not cancel working orders already sitting in MT5, so cancel those separately.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Written by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Reviewed by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Read our full Testing Methodology.

More in this category: Trading Industry News.