The rapid adoption of artificial intelligence in retail trading is creating a visibility gap for broker dealing desks. While automated trading via Expert Advisors (EAs) has been standard for years, the emergence of AI agents using the Model Context Protocol (MCP) allows traders to execute complex strategies that are significantly harder for brokers to identify.
Major platforms have integrated support for this technology. MetaQuotes has introduced native MCP support for MetaTrader 5, and Spotware has launched cTrader AI Agent Connect. These tools allow traders to link external AI models, such as Claude, directly to their live accounts. Once connected, these agents can perform operations like opening or closing positions based on simple natural language requests from the user.
Background
Dealing desks traditionally manage risk by profiling clients based on their trading history and behaviour. They use this data to decide whether to internalise a trade (B-book) or hedge it in the market (A-book). Platforms typically label orders as manual or automated (EA), which helps dealers monitor for specific patterns. EAs generally follow strict technical rules and operate with predictable speed, making them relatively easy to spot. Human traders, by contrast, show more irregular patterns influenced by psychology and market hours. The new generation of AI agents blurs these lines by combining technical and fundamental data, often mimicking human-like, albeit highly disciplined, decision-making.
What it means for traders
For traders, the use of AI agents introduces both potential benefits and operational risks. Because many agents rely on shared models and prompts, they may produce correlated trading patterns, which can lead to rapid, unexpected exposure for the broker. Since these agents can be connected via third-party or open-source servers, there is a risk that sensitive account tokens could be exposed to external providers. If an AI agent makes a mistake due to faulty data or misinterpretation, the lack of clear transaction logs in certain setups may complicate dispute resolutions between the client and the broker. Traders should be aware that brokers are increasingly scrutinising account activity, and the use of unlabelled automated tools may trigger closer monitoring of their trading profiles.