Axiory.ai Launches AI Agent Trading Platform Built on Model Context Protocol

Axiory.ai, a platform developed by the global brokerage Axiory in partnership with Purple Technology, has launched an infrastructure layer designed to allow AI agents to access and trade across multi-asset CFD markets through a single standardised environment.

The platform is structured around four layers: a traditional trading API; a cloud-based code environment for AI workflows; an MCP server enabling direct AI agent connectivity; and a forthcoming strategy builder that will convert natural language trading ideas into autonomous strategies running in the cloud. At launch, the platform supports spot CFDs across FX, equities, and ETFs, with a demo environment available for testing with virtual funds.

Roberto d’Ambrosio, CEO of Axiory, tells TradingTech Insight that the platform is designed to serve traders across the full spectrum of technical ability. “You can use the platform simply to monitor markets, or you can use it to create a trading strategy, place a trade, or manage positions,” he says. “As you grow in both your understanding of AI agents and your market knowledge, you can connect more advanced tools that you’ve developed.”

David Kašper, Director and Co-Founder of Purple Technology, which developed the platform’s technology stack, says the launch reflects a broader shift in how users will interact with financial markets. “Within one to two years, some users will never go to a traditional trading platform interface again,” he says. “They’ll simply tell their agent: ‘Please execute the trades and confirm.’”

MCP as trading infrastructure

In financial markets, however, MCP infrastructure has so far been concentrated on data retrieval rather than execution. Providers including FactSet and LSEG have launched MCP servers for market data access, but the protocol’s use as a conduit for trade instruction and order execution is largely untested. Axiory.ai’s four-layer approach – combining API access, cloud code execution, MCP connectivity, and a natural language strategy builder – represents an attempt to extend the protocol’s reach from data into the execution chain.

Regulatory positioning

The platform’s launch raises questions about how agentic trading fits within existing regulatory frameworks. MiFID II’s algorithmic trading rules and the FCA’s expectations around algo governance both assume either a human or a deterministic system in the decision loop, a framework that does not map neatly onto AI agents interpreting natural language instructions and making contextual decisions about trade execution.

D’Ambrosio says Axiory’s compliance approach rests on positioning the AI agent as a tool rather than a decision-maker. “What we need to ensure is that the AI agent is clearly a tool, just like the indicators, scripts, and automated features already available on trading platforms,” he says. “The client retains full control over the trade: how it’s executed, its limits, and the final decision to execute. That’s what makes this compliant in the current framework.”

He draws a parallel with the automated trading features already embedded in retail platforms. “AI makes this easier because it’s more conversational, but the legal structure behind it doesn’t change. It’s a tool.”

Axiory’s approach also involves a deliberate structural separation between broker and technology provider. “We wanted complete separation, a Chinese wall between what the broker does and what the developers do,” notes d’Ambrosio. Purple Technology builds the product to agreed guidelines, while Axiory controls compliance. “The technology is not controlled by the broker; it’s provided as approved by the broker.”

Risk controls and guardrails

On the question of what prevents an AI agent from executing a catastrophic trade, d’Ambrosio points to two layers of safeguard. Broker-side protections – including zero-balance protection, margin controls, and account-level leverage limits – remain in place regardless of whether the user connects via a traditional platform or an AI agent. “At the end of the day, the connection through an MCP server is essentially a tool,” says d’Ambrosio. “Think of it as plugging in a USB-C port for AI.”

Beyond the broker-side controls, Axiory and Purple Technology have conducted extensive testing of AI agent behaviour in trading contexts. “None of the agents we’ve tested will give you a direct instruction to trade long or short,” notes d’Ambrosio. “We observed their behavioural patterns, and without that understanding, we wouldn’t have moved forward.”

That said, the distinction between deterministic algorithmic systems and LLM-driven agents is material. Traditional algos operate within fixed, rules-based parameters; large language models, by contrast, can hallucinate, misinterpret context, or behave unpredictably under market stress, risks that testing in controlled environments may not fully capture.

CFDs first, exchange-traded instruments to follow

The decision to launch with CFDs rather than direct market access is significant. CFDs are bilateral instruments where the broker acts as counterparty, meaning the systemic risk to broader market infrastructure is more contained than if AI agents were interacting directly with exchange order books. D’Ambrosio confirms that Axiory already offers both CFD and exchange-traded access to its clients, but says the extension to exchange-traded instruments via AI agents will require further work. “When you move to exchange-traded instruments, the way trades are executed and the feedback you receive from the server are fundamentally different. We won’t move there until full, extensive, and satisfactory testing has been completed.”

Kašper outlines a phased product roadmap. The initial launch covers the API connection and MCP server, the lower-risk layers from a regulatory standpoint. A cloud-based strategy builder for CFDs, allowing users to create, backtest, and run strategies, will follow in the coming weeks. Multi-asset expansion via the MCP server is also planned, but Kašper says the timing depends on regulatory coordination with Axiory. “Technically it’s straightforward, but from a regulatory standpoint it’s more difficult,” he says. “That’s why we have strong cooperation with Axiory. They need to tell us when it’s the right time to roll something out.”

Outlook

D’Ambrosio closes with a note of caution. “There can be an impression that everything comes easily,” he says. “We have to make sure that this almost human way of interacting with financial tools doesn’t create new risks.” He points to a broader dependency risk with AI, and emphasised Axiory’s commitment to transparency with clients. “Even though this isn’t a strictly regulated tool, we feel the responsibility to provide full information to our clients.”

Axiory.ai’s launch is one of the first concrete applications of MCP as connective infrastructure between AI agents and live trading environments. Whether the platform gains traction will depend on its ability to navigate the unresolved questions around agentic trading: regulatory clarity on where AI-driven execution sits within algorithmic trading frameworks, the reliability of LLM-based agents in volatile market conditions, and whether the target market – spanning developers and non-technical users – proves to be a coherent product proposition.

What is clear is that MCP’s emergence as a protocol for financial market connectivity is a development that trading technology professionals need to be tracking, regardless of Axiory’s specific outcome. As AI agents become a more established part of the capital markets technology stack, the infrastructure that connects them to market data and execution will become a critical layer of the architecture. And the standards governing that layer are still being written.