Hardly a day went by this past two months without a broker announcing they are allowing traders to connect their app directly to an AI agent. eToro has rolled out Agent Portfolios, ThinkMarkets has ChelseaAI, the Australian division of IG Group now connects directly with ChatGPT, Robinhood launched Agentic Trading, and the list goes on. Crucially, the heavyweights of platform infrastructure have entered the arena. For Ilia Iarovitcyn, CEO of Spotware Systems – cTrader debuted its own version, MCP server for trading, in May – this shift to AI agents is already changing the industry’s distribution layer. “And I believe this is only the beginning,” he notes. “The AI agent will become the primary distribution layer and the main point of interaction between traders and the market. Which doesn't mean trading apps will disappear, but their role will evolve.”For him, in this new world, the trading app will be an execution and data layer, while the AI sits on top, handling the actual interaction with the trader.If traders are now staying within their AI agent ecosystem, then the era of the slick, feature-rich trading app may be entering its twilight. And MetaQuotes is ready to join the trend. Towards a Hybrid Future? The infrastructure making this possible is the Model Context Protocol (MCP), an open-source standard that serves as a universal adapter for artificial intelligence. If the broker is the socket, MCP is the plug that allows any LLM, be it Claude, ChatGPT, or a custom-built bot, to draw data and perform actions.MetaQuotes has been keeping a watchful eye on these developments. The provider tells Finance Magnates that both an MCP solution and an AI Agent integration are coming very soon. Nonetheless, MetaQuotes has not been a walled garden. AI agent integration into MetaTrader 5 (MT5) is already possible through external integration frameworks, backed by technical guides and articles on MQL5.com, providing guidelines and examples for connecting AI systems and trading workflows.This enables AI agents to interact with MT5 across a wide range of scenarios, from accessing market data and executing trades to interacting with Expert Advisors (EAs), with extended control over MT5 client terminal functionality.Public open-source projects and community discussions further demonstrate practical MCP integration strategies for the platform.“On our upcoming MetaTrader 5 release,” says Christoforos Theodoulou, MetaQuotes’ Chief Business Officer, “we are implementing an embedded AI Assistant that acts as an orchestrated coding agent powered by LLM models in the MetaTrader 5 client terminal, allowing users to analyse code, plan multi-step actions, and assist across the workflow, thereby expanding AI agent functionality and capabilities within MetaTrader 5 itself."The assistant will support multiple providers, as well as the provider’s own free solution based on MQL5. It will also operate through MT5's embedded MCP functionality, using either the included internal MCP server or an external MCP server specified by the user.Even so, Theodoulou maintains some skepticism regarding AI agents eventually serving as the predominant gateway for retail access to financial markets."The trading environment is a complex and demanding ecosystem where speed, reliability, security, low latency, compliance and risk management are not merely features but existential requirements. All of these elements must be embedded into every interaction, something autonomous MCP agents may struggle to replicate with the necessary precision and consistency," he says. So, based on this, he expects a hybrid future.“AI agents will serve as an additional access layer for trading services, but not as a replacement for the underlying trading platforms and ecosystems that power global retail trading,” he adds.“The Core Safeguard Is Trade Confirmation, Enabled by Default” The security literature surrounding MCP is already peppered with warnings about tool poisoning, unbounded retrieval and the dreaded infinite loop.An AI agent that misunderstands a prompt might not just give a wrong answer; it might repeatedly retry a disastrous trade or confidently invent a market fact that exists only in its own hallucinatory logic.Iarovitcyn insists that these risks are manageable with the right guardrails. “These are valid concerns, and we took them seriously from the development stage. The core safeguard is trade confirmation, enabled by default,” says the Spotware head. Every action an agent initiates requires a manual thumbs-up from the trader. "Confirmation can be switched off, but only as a deliberate choice, after the trader has had the chance to observe how their agent actually behaves," he adds.Traders can also stress-test their bots in demo environments to catch any digital glitches before real capital is on the line.To manage risks, some are opting for a sandbox approach.Guy Barkat, Director of API Partnerships at eToro, explains that their Agent Portfolios operate as separate accounts. “So they cannot access funds from a user’s main eToro account and are limited to the capital the user has allocated to them,” Barkat said. These agents are also subject to speed limits on how many orders they can execute in a given window, preventing a runaway bot from liquidating a portfolio in a fit of electronic pique.Robinhood’s solution is also sandboxed, as the AI cannot automatically access, view or alter a trader’s primary portfolio or main funds. Others are keeping the leash even shorter. IG’s implementation of MCP is strictly "read-only", allowing the AI to fetch data but forbidding it from pulling the trigger on trades. ThinkMarkets allows for order execution but maintains a firm wall between the AI and the trader’s deposits and withdrawals.While Capital.com’s MCP integration, which is only available for MENA clients, allows the AI agent to access the main account, it requires a two-step confirmation process before making a trade.So, What are AI Agents Really Threatening?“More and more traders interact with markets through AI agents rather than manually: checking positions, reading market conditions, estimating risk, executing,” Iarovitcyn notes.The numbers confirm that retail engagement is beginning to crystallise. A recent survey by Bidget, a crypto exchange, found that 51% of 6000 surveyed already use AI tools to support investment decisions. Robinhood reported that over 50,000 clients opened agentic trading accounts in the first few weeks after the release. In the first few weeks of agentic trading on Robinhood, over 50,000 customers have opened agentic trading accounts and are trading millions of dollars per day of equities and options.Writing and executing sophisticated strategies or optimizing your everyday spending no longer…— Vlad Tenev (@vladtenev) June 18, 2026Still, current MCP implementations largely operate with digital training wheels and true agentic trading remains more theoretical than commonplace. Even so, a shift toward making agents a hub for smart research will alter the face of the industry.And with the two biggest platform providers in the game, the availability barrier has been erased, which indicates an industry approaching a classic innovator's dilemma: despite the threats to adoption, resistance invites obsolescence. Traditional pillars of brand stickiness and the internal cross-selling ecosystems inherent to trading apps will likely need to face a period of radical reappraisal. As Iarovitcyn commented, the app will become a data and execution pipeline. Such a transition may similarly squeeze SaaS-based research tools; as traders bypass trading apps, they will likely consume these analytical capabilities through MCP protocols within their AI agent of choice. Interestingly, MetaQuotes does not consider AI agents to be a threat. “We see them as part of the broader evolution of the industry,” Theodoulou says, emphasising that open protocols alone are unlikely to replace full-featured trading ecosystems in the foreseeable future. “Trusted platforms with mature infrastructure covering all layers of technical and fundamental analysis, algorithmic and social trading, the creation and back-testing of strategies and EAs, VPS and hosting services, risk controls, analytics, marketplace services, liquidity connectivity and aggregation, as well as developer-backed ecosystems, will continue to play a central role regardless of how user interaction layers evolve,” he highlights. This article was written by Adonis Adoni at www.financemagnates.com.