Prediction Market MCP Server (2026): Connect AI Agents to Every Venue

The Short Answer
A prediction market MCP server lets an AI assistant such as Claude or Cursor read live markets, order books and prices by calling tools, with no code from you. Venue-specific servers cover one exchange each. The Predictefy MCP server, @predictefy/mcp, covers 15+ venues through one set of tools: market search, books, candles, trades, cross-venue matching, trader intelligence and paper trading. Live trading tools are there too and are switched on deliberately. Setup is one config entry that runs npx -y @predictefy/mcp with your Predictefy API key.
Key Takeaways
- MCP (Model Context Protocol) is how AI clients like Claude Desktop, Claude Code, Cursor and VS Code call external tools.
@predictefy/mcp1.0.0-beta.11 serves 15+ venues, including Polymarket, Kalshi, Gemini, Limitless, Opinion and Hyperliquid.- Tools cover search, books, candles, trades, matched markets, price comparison, smart money, leaderboards and nine paper-trading tools.
- Live trading tools register with
MCP_ENABLE_TRADE=true; execution requests require a key with thetradescope. Paper tools also need that scope, but do not need the live-trading flag. - It runs on the Free plan, where tool calls draw on 25,000 credits a month.
What is a prediction market MCP server?
An MCP server is a small program that exposes tools to an AI client. Once connected, you can ask "what does Kalshi think about the next Fed decision, and how does Polymarket compare?" and the model calls the server's tools to fetch live data instead of guessing from its training data. A prediction market MCP server is one whose tools read markets, prices and order books, and sometimes trade.
Most prediction market MCP servers wrap a single venue. A cross-venue server lets the model compare the same question across venues in one conversation. For a survey of the options, including venue-specific servers, see our guide to prediction market MCP servers and AI agent tools.
How do you install the Predictefy MCP server?
You need Node.js and a Predictefy API key from the free plan. Then add the server to your client.
Claude Desktop, in claude_desktop_config.json:
{
"mcpServers": {
"predictefy": {
"command": "npx",
"args": ["-y", "@predictefy/mcp@1.0.0-beta.11"],
"env": { "PREDICTEFY_API_KEY": "pk_live_YOUR_KEY" }
}
}
}
Cursor uses the same server object in .cursor/mcp.json. Claude Code adds it from the terminal:
claude mcp add predictefy -e PREDICTEFY_API_KEY=pk_live_YOUR_KEY -- npx -y @predictefy/mcp@1.0.0-beta.11
Pinning the version (@1.0.0-beta.11, current as of October 2026) keeps the tool set stable while the beta line moves; plain @predictefy/mcp follows the latest release. Keep the key out of shared or committed files. Restart the client after editing its config, and the Predictefy tools appear in its tool list.
What tools does it give the model?
| Group | Tools |
|---|---|
| Markets and data | list_venues, search_markets, screen_markets, get_market, get_events, get_orderbook, get_ohlcv, get_trades, get_execution_price |
| Cross-venue | get_matched_markets, compare_market_prices, get_clusters, get_discrepancies, get_arbitrage |
| Trader intelligence | get_smart_money, get_leaderboard, get_wallet_profile, get_market_holders |
| Paper trading | paper_place_order, paper_cancel_order, paper_list_orders, paper_positions, paper_portfolio and four more |
| Live trading (opt in) | exec_quote, exec_prepare, exec_submit, exec_cancel and related tools |
Each tool maps to the same normalized API that Predictefy's SDKs use, so the model sees one record shape whatever the venue. The price-gap and arbitrage tools draw on feeds that are part of the Pro plan.
What can you ask it?
- "Find every open market on the 2028 presidential election across venues and compare the prices."
- "Show me the order book for the leading Polymarket market on the next Fed meeting."
- "Which wallets with high smart-money scores traded on Polymarket in the last day?"
- "Chart the last week of hourly prices for this Kalshi market."
- "Place a paper order for 100 Yes shares at 62 cents and show my paper portfolio."
How does trading through MCP work?
In stages, by design. Out of the box the server registers read, intelligence and paper tools. Paper orders fill against captured real order books, cost 0 credits and never reach a venue, so you can let the model practice freely.
Live trading tools register when you set MCP_ENABLE_TRADE=true and your key has the trade scope. The MCP execution flow builds unsigned artifacts for your own wallet or process to sign. exec_submit previews the request unless confirm is exactly true. Check the venue-specific execution requirements before enabling it. Trading involves risk; start with paper, then small size.
Official references: Predictefy MCP guide.
Frequently Asked Questions
What is the best MCP server for prediction markets?
For more than one venue, the Predictefy MCP server: one set of tools covers 15+ venues, with cross-venue matching, trader intelligence and paper trading included.
Which AI clients does the Predictefy MCP server work with?
Claude Desktop, Claude Code, Cursor and VS Code are documented, and any MCP client can launch it with npx -y @predictefy/mcp.
Is the Predictefy MCP server free?
It runs on the Free plan. Tool calls use credits from the plan's 25,000 a month, and paper trading costs 0 credits.
Can Claude trade prediction markets through MCP?
Yes, once you set MCP_ENABLE_TRADE=true and give the key the trade scope. Orders in this MCP flow are signed on your side. Paper tools do not need MCP_ENABLE_TRADE, but they still require a key with the trade scope.
Which version should I install?
1.0.0-beta.11 is current as of October 2026. Pin it in your config so the tool set does not change underneath you.