Kalshi MCP Server (2026): Connect Claude to Kalshi Markets

The Short Answer
Kalshi runs an official MCP server for its documentation, not for its markets. The endpoint at docs.kalshi.com/mcp gives an AI assistant two tools for searching and reading Kalshi's API docs, which helps when you are writing code. To give Claude live Kalshi markets, order books, prices and trading, you connect an MCP server that wraps the API. Predictefy's MCP server covers Kalshi alongside 15+ other venues with one key, installs in one command, and keeps its trading tools off until you switch them on.
Key Takeaways
- Kalshi's own MCP endpoint searches its API documentation. It does not return live market data or place trades.
- For live Kalshi data in Claude, use an MCP server built on the API, such as Predictefy's
@predictefy/mcp. - One line adds it to Claude Code, and a small JSON block adds it to Claude Desktop.
- Market data, cross-venue comparison and paper trading tools are on by default. Live trading tools stay off until you set
MCP_ENABLE_TRADE=true. - The free plan's 25,000 credits a month cover everyday research, and paper trading costs 0 credits.
Does Kalshi have an official MCP server?
Yes, for its documentation. Kalshi's docs site serves an MCP endpoint at https://docs.kalshi.com/mcp. It exposes two tools: search_api_documentation, which searches the docs and returns matching pages with links, and query_docs_filesystem, which reads a full page. That is useful while you build against the Kalshi API, because your assistant can look up endpoints, fields and examples instead of guessing them.
Adding it to Claude Code takes one command:
claude mcp add --transport http kalshi-docs https://docs.kalshi.com/mcp
What it does not do is serve markets. Ask it for the current price of a Kalshi contract and it can only point you to the documentation for the endpoint that would return it. For live data, you need a server that calls the API itself.
How do you connect Claude to live Kalshi markets?
With an MCP server that wraps a market data API. Predictefy's @predictefy/mcp covers Kalshi along with Polymarket, Gemini, Polymarket US, Limitless and the rest of its 15+ venues, all on one normalized schema, so the same tools work whichever venue you ask about. Create a free API key at portal.predictefy.com/keys, then add the server.
Claude Code:
claude mcp add predictefy -e PREDICTEFY_API_KEY=pk_live_your_key_here -- npx -y @predictefy/mcp
Claude Desktop, in claude_desktop_config.json:
{
"mcpServers": {
"predictefy": {
"command": "npx",
"args": ["-y", "@predictefy/mcp"],
"env": {
"PREDICTEFY_API_KEY": "pk_live_your_key_here"
}
}
}
}
Restart Claude, then check the connection by asking it to list the venues it can see. The list_venues tool answers without calling a venue, so a response confirms the server is running. Then ask for something live, such as "Show me the most active Kalshi markets on the Fed decision and their order books."
What can Claude do with Kalshi through MCP?
| Ask Claude to | Tool it uses |
|---|---|
| Find Kalshi markets on a topic | search_markets or screen_markets with the venue set to Kalshi |
| Read one market or event | get_market, get_event |
| Show the live order book | get_orderbook, or get_orderbooks for many at once |
| Chart price history | get_ohlcv |
| Show recent trades | get_trades |
| Compare a Kalshi price with other venues | compare_market_prices, get_matches |
| Check a gap between Kalshi and another venue | get_arbitrage, which checks both legs against live books |
| Practice a strategy | The paper_* tools, which fill simulated orders against the live book |
Cross-venue comparison is where a multi-venue server earns its place. Kalshi and Polymarket often price the same event differently, and Claude can pull both prices and both books in one conversation, then work out whether the gap survives fees. The arbitrage tool needs the Builder plan or above.
Can Claude trade on Kalshi through MCP?
Yes, if you turn it on. The server registers its ten execution and collateral tools only when you set MCP_ENABLE_TRADE=true, and your Predictefy key needs the trade scope, which you opt into per key. Without both, the server cannot place an order no matter what the conversation asks for.
Trading then runs in steps, and no tool both builds and submits. exec_quote previews the cost against the live book, exec_prepare builds the order, and exec_submit sends it, as a dry run unless you explicitly confirm. Kalshi orders use your own Kalshi API key ID and private key, which Predictefy's execution service uses for that one request and never stores, and your funds stay in your Kalshi account. Kalshi's keys are covered in how to get a Kalshi API key.
A sensible setup is two servers: a read-only one for research, and a separate one with trading enabled that you only start when you mean to trade.
What does it cost to run?
The MCP server is free to install, and its calls use credits from your Predictefy plan. The free plan includes 25,000 credits a month. Market metadata, search and price snapshots cost 1 credit, order book snapshots and historical queries cost 5, cross-venue comparisons cost 10, and an arbitrage query costs 15. Paper trading costs 0 credits on every plan.
What about other Kalshi MCP servers?
Several community servers on npm and GitHub wrap Kalshi's public API, from read-only data wrappers to ones that place orders. Evaluate them like any other dependency: when the last commit landed, whether there is a licence, how credentials are stored, and whether an unsupported request returns an honest error or an empty result. A Kalshi-only server also answers only Kalshi questions. The moment you want to compare a price with Polymarket or Gemini, you are running several servers and reconciling their output by hand.
For the Polymarket side, see how to connect Claude to live Polymarket markets, or start with the Predictefy API.
Frequently Asked Questions
Does Kalshi have an MCP server?
Kalshi runs an official MCP endpoint at docs.kalshi.com/mcp for its documentation, with tools to search and read the API docs. It does not serve live markets or place trades. For live Kalshi data, connect an MCP server built on the API, such as Predictefy's.
How do I connect Claude to Kalshi?
Create a free Predictefy API key, then run claude mcp add predictefy -e PREDICTEFY_API_KEY=your_key -- npx -y @predictefy/mcp in Claude Code, or add the same server to claude_desktop_config.json for Claude Desktop. Claude can then search Kalshi markets, read order books and compare prices with other venues.
Can Claude place trades on Kalshi?
Only if you enable it. Predictefy's MCP server keeps its execution tools off unless you set MCP_ENABLE_TRADE=true and use a key with the trade scope. Kalshi orders then use your own Kalshi API key ID and private key, which are used for one request and discarded.
Is a Kalshi MCP server free?
Kalshi's documentation MCP is free, and Predictefy's MCP server is free to install, with calls drawn from your plan's credits. The free plan includes 25,000 credits a month, and paper trading costs 0 credits.
Does the Kalshi MCP server work with Claude Code?
Yes. Add Kalshi's documentation server with claude mcp add --transport http kalshi-docs https://docs.kalshi.com/mcp, and Predictefy's live data server with claude mcp add predictefy followed by your key and npx -y @predictefy/mcp.