Kalshi Screener MCP

Kalshi Screener is an independent project. It is not affiliated with, endorsed by or sponsored by Kalshi. Market information only, not financial advice.

Screen Kalshi prediction markets from your AI app. Ask things like:

• "Find markets priced 80c or higher that close in the next 48 hours."

• "Which markets were listed in the last day?"

• "Show every market in this event ranked by price."

• "Run my 'Closing soon' screen and list today's alerts."

Works with Claude (Desktop, Code and web), ChatGPT, Grok, Meta Muse, Cursor and VS Code. It pairs with the free Kalshi Screener Chrome extension: your saved conditions, watchlist and alerts are shared with the AI, and every result has an Open in Kalshi Screener link.

Read-only. It uses Kalshi's public market data, never places or cancels orders, and gives no buy or sell advice. No Kalshi account or API key is needed.

No hosted service. The server runs on your computer or on your own Cloudflare account. Nothing goes through the author's servers.

1. Desktop AI apps (runs on your computer)

Requires Node.js 20 or later. The server is not published to npm; it runs from this folder.

1. Download this repository (on GitHub: Code, Download ZIP) and unzip it.

2. Open a terminal in the unzipped folder and run npm install once.

3. Point your AI app at bin/cli.js in that folder. Replace C:/path/to/kalshi-screener-mcp below with your folder's full path (forward slashes work on Windows too).

Claude Desktop, Cursor, VS Code and other MCP apps. Add this to the app's MCP settings (in Claude Desktop: Settings, Developer, Edit Config):

{
  "mcpServers": {
    "kalshi-screener": { "command": "node", "args": ["C:/path/to/kalshi-screener-mcp/bin/cli.js"] }
  }
}

Claude Code:

claude mcp add kalshi-screener -- node "C:/path/to/kalshi-screener-mcp/bin/cli.js"

Restart the app. In the Chrome extension, Settings, AI Agents (MCP) builds the same settings from the folder path you paste and has a copy button.

Share data from the Chrome extension

In the extension, open Settings, AI Agents (MCP), turn on Share with the MCP server on this computer and press Check and share now. The extension then sends your condition presets, watchlist and recent alerts to 127.0.0.1 whenever they change and after each background check. API keys and AI settings are never sent. Only the extension can write to this local endpoint; web pages are refused.

The first extension that shares data is paired with the server, so other installed extensions cannot overwrite your data. If you reinstall the extension and it gets a new ID, start the server once with --pair (for example node bin/cli.js --pair in the server folder, or add "--pair" to args and remove it afterwards).

Open in Kalshi Screener links use chrome-extension://. Some desktop apps cannot open them directly; copy the link into Chrome's address bar instead.

Related MCP server: Prediction Markets MCP Server

2. Web AI apps (ChatGPT, Grok, Meta Muse, Claude on the web)

These apps connect to a web address, so deploy your own free copy to Cloudflare Workers.

1. Create a free Cloudflare account.

2. Deploy this folder. In the unzipped folder, run:

   npm install
   npx wrangler deploy

   Wrangler creates the storage it needs on first deploy.

3. Choose a long random access key and save it:

   npx wrangler secret put ACCESS_KEY

4. Add the server to your AI app as a custom connector or MCP server:

   https://kalshi-screener-mcp.<your-subdomain>.workers.dev/mcp?key=<your access key>

   • ChatGPT: Plugins, Add custom MCP server, paste the URL.

   • Claude on the web: Settings, Connectors, Add custom connector, paste the URL.

   • Meta Muse: tell Muse the server URL in chat and store the access key when it asks.

   • Grok (xAI API): pass the URL as a remote MCP tool, or send the key as Authorization: Bearer <key>.

Optional: set EXTENSION_ID (npx wrangler secret put EXTENSION_ID) to your extension ID so remote results include Open in Kalshi Screener links.

Cloudflare's free plan allows about 50 outbound requests per call. Very broad searches over every open market can exceed that; the server then asks you to narrow the search (for example scope: "closing_24h"). The local server has no such limit.

Tools

Prices are dollars from 0 to 1, the same as the implied probability of YES. Scopes: open_markets (default, excludes multivariate combo markets), closing_24h, closing_7d, multivariate_only.

Historical evidence

Ask your AI "Will this go up?" or "Is this likely?" about a market and it can call get_market_evidence. The tool looks at the last 20 settled events in the same Kalshi series (up to 40) and reports:

• Similar price history: in past events, markets priced within 5c of this market's current price at the same time before close, and how many of them settled YES. One market per past event is counted, so related outcomes are not counted twice.

• Favorite record: how often the highest priced market in each past event won.

• Recent events: what won, what the favorite was, and the prices at the time.

Every rate comes with its sample size and a 95% range (Wilson score interval), plus a confidence label: very_low under 10 comparable events, low under 30, medium under 100. Prices are the mid of the YES bid and ask when the spread is 10c or less, otherwise the last trade. Older events come from Kalshi's /historical endpoints.

This is a summary of how a series settled in the past. It is not a prediction, and a small sample says little. Each event can take one Kalshi request, so on a free Cloudflare Worker a long history may stop early; the result says so in data_notes, and the local server has no such limit.

Add browser research (optional)

Most AI apps can search the web on their own. News sites, X and forums often need a login or block bots. To let a desktop AI app read those pages in your own signed-in Chrome, add Microsoft's free Playwright MCP next to this server:

1. Install the Playwright MCP Bridge Chrome extension.

2. Use these MCP settings (the extension's Research with AI window has a copy button):

{
  "mcpServers": {
    "kalshi-screener": { "command": "node", "args": ["C:/path/to/kalshi-screener-mcp/bin/cli.js"] },
    "playwright": { "command": "npx", "args": ["@playwright/mcp@latest", "--extension"] }
  }
}

3. Ask, for example: "Find markets over 80c closing within 24 hours, then research the top one with its research_prompt."

LM Studio (0.3.17 or later) reads the same format from %USERPROFILE%/.lmstudio/mcp.json on Windows and ~/.lmstudio/mcp.json on macOS and Linux, so local models can use both servers. Other options: Claude in Chrome for Claude paid plans and ego lite for Claude Code, Codex and Cursor. These are third-party tools, not affiliated with Kalshi Screener. They can read any site you are signed in to, so only connect accounts you are comfortable letting the AI read. The research prompt tells the AI to read only and never post or trade.

Command line options

node bin/cli.js            MCP over stdio (default)
node bin/cli.js --http     MCP over HTTP at http://127.0.0.1:47811/mcp
  --port <n>       Port for the extension bridge and --http (default 47811)
  --state <path>   State file (default ~/.kalshi-screener/mcp-state.json)
  --no-bridge      Do not accept data from the Chrome extension
  --pair           Pair with the next extension that shares data

KALSHI_SCREENER_EXTENSION_ID sets the extension ID used in links before the extension has shared data.

Questions and inquiries

This project does not take code contributions. For bugs, open an issue. For advertising in the Chrome extension, open an issue titled "Advertising inquiry".

License

Free for individuals, including your own trading, research and analysis. Business use needs a commercial license.

Licensed under the PolyForm Noncommercial License 1.0.0 with an additional permission for individuals' personal trading. Companies, funds, paid services, bundling into other products and paid signals need a separate commercial license: open an issue titled "Commercial license" in this GitHub repository. See LICENSE for the full terms.