A trading-account connector should begin as a way to inspect information, not as a reason to trust an assistant with decisions. Start with the narrowest access offered, compare what the assistant says with the broker's own account view, and inspect every proposed action before approving it.
That distinction matters in eToro's October 8, 2026 announcement of connectors for Claude, ChatGPT, Grok and Grok Bot. The company says the MCP-based connection can bring account context into those assistants without API keys or code. Wall Street en Español independently reported the launch and described the same access boundaries (eToro's announcement; Wall Street en Español's coverage). This is a review of announced capabilities, not a hands-on test of the connector.
What account access does the connector announce?
The announcement says a connected assistant can answer questions using account holdings, positions, balances and performance. It describes two modes: read-only access, and read-and-act access where available. In the latter mode, an action is proposed for the user to review and approve; funding and withdrawals remain inside eToro. Users can also pause or disconnect the connection. Wall Street en Español reports the same distinctions.
Availability is not universal: eToro says capabilities depend on the assistant and jurisdiction, and that the connectors were not available to clients in the US, Australia or Singapore at launch. Check the current terms for your own account and assistant rather than assuming the launch description applies everywhere.
A practical check before you connect
Use this sequence to inspect the boundary; it does not certify the service or the quality of an AI answer.
| Check | What to verify |
|---|---|
| Confirm the exact assistant and account | Check the provider's current availability and the account you are authorising. Do not infer access from an assistant's name or a connector directory listing. |
| Begin with read-only access | If that option is available, use it first. Ask a simple question about information you can independently see, then compare the answer with the broker's own account screen. |
| Treat an answer as a summary, not a ledger | If positions, balances or performance do not match the broker view, stop and resolve the discrepancy there. Do not use a conversational answer as the record of the account. |
| Inspect any proposed action | For a read-and-act workflow, check the instrument, direction, size and order details shown to you. Approve only an action you independently intended; a fluent explanation is not evidence that the details are correct. |
| Locate the pause and disconnect controls | Know where to stop the connection before using it. The announcement says these controls exist; this checklist does not establish how quickly access is revoked in every assistant. |
The last checks follow a broader security principle, not a claim that eToro has a specific vulnerability. OWASP's GenAI guidance describes prompt-injection risks and recommends least privilege and human approval for high-risk actions. The MCP project's own security guidance also discusses authorization and consent failures. Neither document audits this particular connector (OWASP GenAI Security Project; MCP security best practices).
What approval does not settle
An approval step is a boundary to inspect, not a guarantee that an assistant understood your request or that a resulting action will match what you expected. eToro's announcement says AI-generated outputs are not warranted for accuracy and warns that systems can act quickly or be difficult to stop in real time. It also tells users to verify positions and fees in the account.
So keep the broker's account view as the source of record. If you use a connector, begin with read-only questions and reconcile the answer yourself. If you later enable actions, review the proposed order itself rather than the assistant's summary of it. Our guide to trading-AI permissions in cTrader covers the separate question of how an automation workflow should be bounded; reviewing AI-generated cBot code covers checking changes before they reach an account.
The useful test is not whether an AI can talk about your portfolio. It is whether you can see, limit and verify exactly what account information and actions sit behind that conversation.