// GLOSSARY

MCP (Model Context Protocol)

An open protocol that lets AI models use external tools and data through a standard interface. An MCP server exposes typed tools (e.g. search_markets, place_order) that clients like Claude can call — the standard way to give an AI agent trading capability.

The Model Context Protocol standardizes how AI applications connect to outside capabilities. Instead of bespoke plugins per model, a server declares its tools — names, descriptions, JSON schemas — and any MCP-capable client (Claude, Claude Code, and a growing list of others) can discover and call them.

Why traders care

An MCP server in front of a trading API turns "AI that can talk about markets" into "AI that can act in them": the model calls search_markets, inspects the result, and calls place_order, all within one conversation. The protocol handles the plumbing; the server decides what's allowed.

The design question

Because the model chooses when to invoke tools, the server is the safety boundary — it should enforce risk limits rather than trust the model's judgment. Mithril's hosted MCP server (mcp.trymithril.com/mcp) exposes search_markets, place_order, cancel_order, and get_portfolio for Kalshi and Polymarket, with your notional caps and kill switch enforced server-side. A trader-oriented introduction: what is MCP for traders.