Not a demo. Not a chatbot answering price queries. A full workflow handling inventory management, dynamic quoting, position sizing, execution, monitoring, and Telegram alerts. The kind of thing that normally requires a quant, a developer, backtesting infrastructure, and weeks of iteration.
Two engineers at Revolut X just mass-produced a question that should keep every fintech product leader up at night: if an AI agent can orchestrate your entire system through the API, what exactly is your product team building?
a blog by Linas Beliunas, Country Manager, Europe, and General Manager, Flutterwave Lithuania
Nikita Ivanov and Vlad Kaminski, engineers on Revolut’s crypto exchange, connected Claude to the Revolut X API via Anthropic’s Model Context Protocol (MCP) as a side project. Leonid Bashlykov, Revolut’s Head of Crypto Product, then prompted Claude in plain English and said it had a working market-making strategy running in roughly 30 minutes.

Let that sink in.
MCP is the load-bearing detail here. Anthropic released it in November 2024 as an open protocol that lets AI models discover and use external tools through a standardized interface.
Before MCP, every AI integration was bespoke plumbing. Now, a fintech can stand up one MCP server, and any compatible agent can trade, pull data, run backtests, and set alerts against it.
In other words, Revolut built the server; Claude figured out how to use it.
The uncomfortable implication here is less about trading and more about product strategy. Revolut’s own team is now openly asking whether traditional feature roadmaps – design a screen, add a button, ship an update – still make sense when an agent can compose capabilities on its own.
The API surface becomes the product. The UI becomes a monitoring layer.
What MCP unlocks that a standard API never could
According to Leonid Bashlykov:
🧠 Agents that reason, not just execute – A bot follows rules. An MCP-connected agent reads your portfolio, checks conditions, pulls context – and decides.
📊 Sophisticated strategies in plain language — „Rebalance to 60/40 BTC/ETH if BTC dominance drops below 52%.” That’s now a prompt. The barrier to complex strategies just collapsed.
🔄 Full trading lifecycle, not just order placement – Screening → sizing → timing → execution → monitoring → alerting. One connected workflow, not five separate scripts.
🔗 Composability with anything – Trade AND read news AND track on-chain flows AND update a dashboard. MCP is the connective tissue. Revolut X is the execution layer.
Leonid Bashlykov concluded: „Tbh, most „MCP for trading” talk right now is still toy stuff — price alerts, simple queries, basic automations. That’s not where this ends up. The question is to which point it continues to make sense bringing UI updates to the product if AI enabled flow allows so much more capabilities. Follow what comes next 👀”
What’s next?
Linas Beliunas commented: „Looking ahead, watch for two things in the next 12 months. First, whether Coinbase, Kraken, and Interactive Brokers ship their own MCP layers. If they do, the competitive moat moves from interface design to API richness and execution quality. Second, who builds the permission and risk guardrails that make agent-driven trading palatable to regulators. Crypto is a forgiving sandbox – 24/7 markets, fewer legacy rules. Equities won’t be. The fintech that solves scoped agent permissions (”trade up to $10k/day, BTC and ETH only, kill-switch at 2% drawdown”) for regulated markets will own a very large category. Revolut fired a starting gun. The race is infrastructure, not interfaces. Paradigm shift.„
Banking 4.0 – „how was the experience for you”
„To be honest I think that Sinaia, your conference, is much better then Davos.”
Many more interesting quotes in the video below: