AI & Agents
Prop Trading
AI trading agents query real-time implied volatility surfaces, term structures, and Greeks natively via Model Context Protocol. Autonomous agents monitor vol regime shifts, identify mispricings, and execute systematic strategies without custom API integration.
Institutional crypto trading is moving toward automation. Systematic strategies, risk monitoring, and research workflows increasingly rely on AI agents that can query data, evaluate conditions, and act — without a human in the loop for every data request.
The bottleneck has been data access. Traditional REST APIs require custom integration: authentication flows, endpoint mapping, response parsing, error handling. Each data provider means another integration to build and maintain. Model Context Protocol (MCP) removes this by providing a standardised interface that AI agents query natively — no per-provider custom code.
MCP is not a one-way data pipe. The protocol supports bi-directional communication between the AI agent and the data server. An agent can refine its queries based on intermediate results — requesting a broad instrument scan, then drilling into specific strikes based on what it finds, then backtesting a strategy using those exact instruments. The server responds contextually, and the agent adapts in real time.
This elicitation capability means the agent doesn’t need to know the exact query upfront. It can explore the data space conversationally, narrowing from broad to specific as the analysis develops — the same way a human analyst would work through a research question, but at machine speed.
Real-time and historical data across the full derivatives stack. Implied volatility surfaces, options Greeks, funding rates, open interest, term structures, and index prices across BTC, ETH, SOL, and additional assets. Handles spot, perpetual, futures, and options data from 22+ exchanges in a single tool call.
Current instrument listings across exchanges, including expiry dates, strike prices, and contract specifications. An agent can programmatically identify which contracts are trading before querying their data — no hardcoded instrument lists, no stale mappings.
Systematic strategy backtests using historical data. Define entry and exit conditions, specify instruments (spot, perpetual, futures, or options), and receive P&L results within a single tool call. Agents can iterate on strategy parameters without leaving the context window.
An agent queries the 30-day and 7-day ATM implied volatility term structure hourly. When the spread inverts beyond a threshold, it triggers an alert or initiates a pre-defined strategy — pulling the exact surface points needed to evaluate the trade.
An agent queries IV surfaces across multiple exchanges via the composite data feed, comparing venue-specific pricing to the exchange-weighted aggregate. Persistent divergences flag potential arbitrage opportunities for review or automated execution.
An agent pulls current market data — vol surfaces, funding rates, open interest by strike — and generates structured market commentary grounded in real numbers. No boilerplate, no stale data, no manual data extraction.
For teams building with AI agents, MCP versus traditional REST comes down to development velocity and analytical depth. MCP tools are discoverable by the AI model with no wrapper code or SDK dependency. Structured error handling operates at the protocol level. Composability means an agent can chain data retrieval, instrument discovery, and backtesting in a single conversation based on intermediate results. And the bi-directional nature of the protocol means the agent gets smarter with each query — refining its analysis based on what the data reveals, not what the developer anticipated.
The MCP server provides the same data available through the REST and WebSocket APIs: SVI-calibrated implied volatility surfaces, options Greeks, funding rates, open interest, mark prices, index prices, and settlement prices. Data spans spot, perpetual, futures, and options across Deribit, Bybit, OKX, GateIO, and additional major exchanges. Historical and real-time data are accessible through the same tool interface with the same data quality guarantees.
