TweedBeetle/prediction-market-interface

Kalshi MCP Server

FastMCP-based server enabling Claude and other LLMs to interact with Kalshi prediction markets through natural language.

What This Is

A Model Context Protocol (MCP) server that provides LLM agents with tools to:

• Research prediction markets
• Execute trades (market & limit orders)
• Manage portfolios and positions
• Monitor account activity

Built with FastMCP for seamless integration with Claude Code, Claude Desktop, and other MCP clients.

Current Status

See for complete status, roadmap, and implementation details.

Key Features

• 🔐 Secure Authentication - RSA-PSS signature-based API authentication
• 🌍 Dual Environments - Separate demo (test money) and production (real money) servers
• 🛡️ Safety Limits - Built-in validation and configurable order size limits
• 📊 Rich Data Models - Type-safe Pydantic models with LLM-friendly interpretations
• ✅ Comprehensive Testing - 60+ integration tests with VCR cassette replay

Quick Start

Setup Instructions: See

Usage Example:

User: "@kalshi_demo search for bitcoin markets"
User: "@kalshi_demo check my balance"
User: "@kalshi_demo buy 10 YES contracts on KXBTC-31DEC-50K at 45¢"

Documentation

• - Complete setup guide
• - Development patterns and technical reference
• - Product requirements, status, and roadmap
• - Testing infrastructure

Project Structure

src/kalshi/
├── kalshi_mcp_server.py    # FastMCP server with 16 tools
├── client.py               # Kalshi API client
└── models.py               # Pydantic models

run_kalshi_mcp_demo.py      # Demo wrapper (test money)
run_kalshi_mcp_prod.py      # Production wrapper (real money)
tests/                      # 60+ integration tests

Requirements

• Python 3.11+
• uv package manager
• Kalshi account with API access (demo or production)

Security

All credentials are gitignored:

• .env.kalshi.demo / .env.kalshi.prod
• secrets/ directory with private keys

See for secure credential management.

License

See LICENSE file.