Unhedged adds AI assistant connections for prediction-market trading

Unhedged has released tools intended to let users research and trade on its prediction-markets platform through compatible AI assistants, while giving developers a TypeScript interface for building their own research and trading applications.

The Oct. 22 release paired an updated Model Context Protocol, or MCP, server, version 0.3.1, with version 0.1.0 of its TypeScript software development kit. Unhedged covers markets tied to esports, cryptocurrency, sports and other real-world events.

Through Unhedged MCP, users can connect accounts to Claude Desktop, Claude Code, Codex, Cursor and other compatible assistants. In a conversation, an assistant can search markets, inspect resolution rules and review a user’s existing positions. With trading permission enabled, it can also place and manage orders.

A sample workflow offered by Unhedged is a request to identify active Bitcoin markets, explain the criteria for resolving them and retrieve the user’s related positions. The information is drawn from Unhedged into the AI conversation.

The MCP documentation recommends that users start with read-only access before enabling trading. Permissions can be restricted to market, position and balance research; expanded to account data, bets, orders and cancellations; or extended to trading and fund transfers.

The server can search active and ended markets by asset, sport, topic or tag, and retrieve outcomes, closing times and resolution rules. It also supports reviews of balances, positions, payouts, orders, trades, quotes and account events. Its trading functions include limit, market, stop-loss and take-profit orders, batch orders, cancellations and replacement of open limit orders.

Unhedged says access is governed by scoped API keys, with account information and trading functions requiring the corresponding permissions. A session spending cap can restrict stakes and certain order commitments in U.S. dollars, although it resets when the MCP server restarts. The company also says individual positions, trade sizes and activity are not visible to other users.

For developers, the TypeScript SDK supplies a typed client for research tools, portfolio dashboards and trading integrations. It handles HTTP requests and provides TypeScript definitions for inputs and responses, including autocomplete and clearer information on expected returned data.

Public market data can be queried without an API key, while account and trading methods require scoped keys. Optional WebSocket clients provide public order-book snapshots and an authenticated account-event stream; the event client tells an application when to refresh account state through REST. The SDK also surfaces uncertainty or partial completion in trading-operation errors and does not automatically retry mutations.

The package, @unhedged/typescript-sdk, is available through npm for Node.js 20 or later projects using ESM.

MCP is a standardized translation layer that can connect AI systems with external services, including market data, account details and trade execution, according to a Forbes Councils article. Alpaca said its MCP server similarly enabled natural-language market research, analysis and order placement across AI interfaces and coding tools.

Unhedged warned that prediction markets involve substantial risk of loss and that participation is subject to age and jurisdiction requirements in its terms of service.