Kalshi published a hosted MCP connector, so an AI assistant can now read its markets, prices, rules and orderbooks straight from the exchange, and read your portfolio once you connect an account. That's a good thing for traders. It also draws a clean line around what one venue's server can do, and the rest of this piece is about that line.
PredictionMarketsPicks is not affiliated with or endorsed by Kalshi. We run an independent MCP server, and it's meant to sit alongside Kalshi's, not replace it.
What Kalshi's server covers
Everything in this section comes from Kalshi's own connector documentation, read on Nov. 6, 2026. The endpoint is https://api.elections.kalshi.com/mcp, over Streamable HTTP.
| Group | Tools | Needs an account? |
|---|---|---|
| Market data | search_markets, get_event_odds, get_market, get_market_orderbook, get_market_links | No |
| Your portfolio | get_balance, get_positions, get_orders, get_fills, get_settlements | Yes, read permission |
| Perps | Market, orderbook and link tools, plus balance, positions, orders, fills and risk | Market tools no; account tools yes |
| Price alerts | get_price_alerts, create_price_alert, delete_price_alert | Yes |
Three details are worth knowing. Account access runs through OAuth, so you never paste an API key or private key into your assistant. The orderbook tool reads up to 10 levels per side. And the connector doesn't trade for you: its link tools return a market link, and you review and submit any order on Kalshi.
If your question is "what does Kalshi say about this market" or "what do I hold," that server is the right answer.
What a single-venue server can't do
The limit is structural. An exchange's server describes that exchange, so four questions traders actually ask sit outside it:
- Is this cheaper somewhere else? That needs the same contract priced on other venues, after each venue's own fee.
- Is this combo quote fair? A combo price depends on how the legs move together, which is a modeling job, not a market-data read.
- How much should I size? Sizing needs your probability against the price, and that's math on top of the data.
- Has this signal actually worked? That takes a graded record kept by someone other than the venue.
That's the layer we built. PredictionMarketsPicks is the independent quant layer for prediction markets: fee-adjusted edge, Kelly sizing, correlation-aware combo pricing and best price across Kalshi, Polymarket US and Novig, with a graded public record. The server carries 46 tools in the registry (40 listed on the main endpoint right now).
How to run both MCP servers
Both are remote servers, so a client that takes a standard MCP config can list them side by side:
{
"mcpServers": {
"kalshi": { "url": "https://api.elections.kalshi.com/mcp" },
"predictionmarketspicks": { "url": "https://predictionmarketspicks.com/api/mcp/mcp" }
}
}Hosted apps like Claude and ChatGPT add each one as a custom connector instead. Our setup page has the steps for each client, and the comparison page sets the servers side by side. Kalshi says some apps will ask for client credentials during account connection, and its documentation has a connection help section for that.
Three prompts that use the second server
These are real calls to our tools. The first is a live read, and the other two are worked examples with made-up inputs.
1. "What's the cheapest way to buy the Bears to win this Sunday?" That's best_price. As of 8:00 AM ET on Nov. 6, the cost to buy one Bears contract against Tampa Bay, fee included, was:
| Venue | Cost per contract, after fees |
|---|---|
| Novig | 71.5¢ |
| Kalshi | 72.4¢ |
| Polymarket US | 72.7¢ |
The table is ranked by price only, and prices move. Novig was the cheapest at that moment by $0.94 per 100 contracts. Kalshi's server can give you the middle row, and the other two are why you'd run the second one. Polymarket US prices are shown for information. PolymarketPartner: PredictionMarketsPicks is a paid Polymarket partner.
2. "Is this combo quote fair?" That's combo_edge. Say two legs price at 68¢ and 58¢, you put the chance both hit at 44%, and the quote is +180. The tool returns a fair price of +127, a fair band of +114 to +142, and an expected value of +23.2% at the quote. It works on any combo price you can paste in, Kalshi's or anyone else's.
3. "How much would I put on it?" That's kelly_size. With a 44% estimate, a 36¢ price and a $1,000 bankroll, quarter-Kelly comes to $31, or 3.1% of the bankroll.
Kalshi's server or ours: the short version
Kalshi's server is the primary source for Kalshi. Add ours when the question crosses venues, involves a combo, or needs a size and a record behind it. Both are free to connect, and you can start with the MCP hub or the Kalshi setup guide.
Trade responsibly.