Kalshi Multi Agent Research Server

A Model Context Protocol (MCP) server that exposes Kalshi prediction market data as tools for Claude Cowork. Designed to power a multi-agent research pipeline where specialist sub-agents (trend, orderbook, news, base-rate, fundamentals) independently analyze markets and produce confidence-rated trade signals.

Architecture

• Python MCP server built on FastMCP (streamable HTTP transport)
• Read-only Kalshi API integration (RSA-PSS signed requests)
• Deployed as a remote Claude connector via ngrok tunnel
• Orchestrated by a Cowork prompt that spawns parallel specialist agents

Tools exposed

• list_markets — filter open/closed/settled markets
• get_market — detail on a single market
• get_orderbook — bid/ask depth
• get_recent_trades — recent fill history
• get_candlesticks — historical OHLC data
• web_research — scrape external context

Setup

1. Install dependencies: uv sync
2. Create .env with KALSHI_KEY_ID and KALSHI_PRIVATE_KEY_PATH
3. Place your Kalshi private key at the path specified in .env
4. Run: uv run python server.py
5. Expose via ngrok, add as a custom connector in Claude Desktop

Disclaimer

Research tool only. Not financial advice. Prediction markets carry real financial risk; AI-generated confidence scores are one signal, not a recommendation to trade.