What is a Kalshi MCP server?
A Kalshi MCP server translates Kalshi's exchange APIs into tools an AI client can discover and call. The server sits between the model and Kalshi. It can expose public market data, private portfolio data, or live order actions depending on its code and the credentials you provide.
Kalshi does not need MCP to support programmatic trading. Its official documentation covers REST, WebSocket, and FIX interfaces for real-time market data and execution. MCP adds a standard tool layer so an agent can request an order book, inspect settlement rules, check positions, or prepare an order through natural-language instructions.
The distinction matters. Kalshi owns the exchange API. Community maintainers currently own the MCP servers visible in search. Review the repository, permissions, dependencies, and release history before giving any server access to an account.
Start with the Kalshi API guide if you need the exchange layer before the agent layer.
Is there a Kalshi MCP server I can connect to Claude?
Yes. Multiple community-built Kalshi MCP servers can connect compatible AI clients to market data and trading functions. The strongest current implementations publish their source code, list every exposed tool, document credential handling, and separate read-only research from order actions. None of the repositories reviewed here is an official Kalshi product.
The 9crusher implementation exposes market discovery, settlement rules, order books, candles, trades, balances, positions, fills, settlements, and order management. It defaults to Kalshi's demo environment and requires an explicit confirmation value for order creation and amendment. Its public repository also documents offline tests for order translation and tool registration.
The BrainDAO implementation exposes market discovery, pricing, portfolio, and trading operations through an npm package. Its repository says the server runs locally and sends requests only to Kalshi's official API. Both projects require careful review before production use because an MCP tool with trading permission can move real capital.
For the broader automation stack, see the Kalshi bot guide.
What can a Kalshi MCP server do?
A trading-capable Kalshi MCP server can give an agent four practical tool groups: discovery, research, account state, and execution. Exact coverage varies by repository. Traders should inspect the tool list and test each action against Kalshi's demo environment before deciding that a server is ready for a live account.
Discovery tools list markets, events, and series. Research tools read order books, recent trades, candles, exchange status, and contract rules. Private account tools return balances, positions, fills, orders, and settlements. Execution tools create, amend, reduce, or cancel orders.
That tool surface gives an agent access. It does not supply a model, an edge, or a risk policy. The agent still needs explicit limits for market selection, order size, price tolerance, stale data, open exposure, and failure handling. A prompt cannot replace those controls.
Read prediction market API options before choosing between a native integration and a unified data layer.
- Discovery: markets, events, series, and exchange status
- Research: rules, order books, candles, and recent trades
- Portfolio: balances, positions, fills, orders, and settlements
- Execution: create, change, reduce, and cancel orders
What do I need besides an LLM to trade Kalshi automatically?
An LLM needs a controlled execution stack around it: a Kalshi account, API credentials, an MCP client, a reviewed server, deterministic risk checks, logs, reconciliation, and a kill switch. Keep the private key outside prompts and model context. Run the server locally or inside infrastructure you control.
MCP's official specification treats tool execution as a security boundary. It calls for clear user consent, access controls, visible tool activity, and a human who can deny invocations. That guidance carries extra weight when a tool can create an order. Tool descriptions and model output are inputs, not authorization.
Start read-only. Record the agent's proposed trades. Compare those proposals with the live order book and the contract's settlement rules. Move to Kalshi's demo environment next. Live access comes last, with small limits and a confirmation gate that the model cannot bypass.
Event contracts carry market risk. Software can submit the order you asked for and still lose the trade.
When should a trader use Kairos instead?
Kairos fits traders who want cross-venue data and execution without maintaining an agent server. It puts Kalshi, Polymarket, and Predict.fun in one book with sub-second data, global best bid and best ask, advanced order types, and low-latency execution. The trader keeps direct control of the workflow.
Use a Kalshi MCP server when the strategy depends on an AI agent, custom research, or proprietary automation against Kalshi's API. Use Kairos when the edge comes from seeing fragmented liquidity and executing across venues from one terminal. The prediction market terminal comparison covers that decision in detail.
Build the agent when the agent creates the edge. Open Kairos when the trade does.
Sources
This guide uses Kalshi's API documentation for official API coverage, the Model Context Protocol specification for the protocol and safety model, and the public repositories for the 9crusher Kalshi MCP server and BrainDAO Kalshi MCP server for current tool and configuration claims. Repository capabilities can change. Review the live source before installation.