Skip to main content
Glama
Kalshi Eye MCP Server

Kalshi Eye MCP Server

by KalshiEye

Kalshi Eye MCP server

An MCP server that lets AI assistants such as Claude and Cursor search the Kalshi Eye archive of prediction market data. It can find Kalshi, Polymarket and Limitless markets, show how a market's price moved, check what was recorded, and list the order book files you can download.

It needs a Kalshi Eye API key: create one at API keys (a free account works) and set it as KALSHIEYE_API_KEY.

Install

Claude Code:

claude mcp add kalshieye -e KALSHIEYE_API_KEY=clab_your_key -- uvx kalshieye-mcp

Claude Desktop, Cursor and other clients (claude_desktop_config.json, .cursor/mcp.json, ...):

{
  "mcpServers": {
    "kalshieye": {
      "command": "uvx",
      "args": ["kalshieye-mcp"],
      "env": { "KALSHIEYE_API_KEY": "clab_your_key" }
    }
  }
}

Without a key, or with a placeholder, every tool answers with how to get one. Listing files and download links cover the datasets your plan includes.

uvx comes with uv. With pip instead: pip install kalshieye-mcp, then use kalshieye-mcp as the command.

Related MCP server: Polymarket MCP Server

Tools

Tool

What it does

search

Find series, events and markets by keyword, venue, category or status. Series lead and come a page at a time (series_total, has_more, offset); events and markets are the 10 best whose words hold the query, with events_found / markets_found (exact up to 10,000, then *_found_exact: false) and more_events / more_markets, the paged calls that list every match

get_series

One series: what it is, how often it runs, markets in the catalog, and days of data files

list_markets

A series' markets, or the markets on every venue holding each word of q (venue narrows it), newest first, with status and result. A page at a time: total (with q, exact up to 10,000, then total_exact: false), and next_cursor (with q) or next_offset, passed back until it is null

list_events

Every event on every venue whose title, series name or category holds each word of the query, newest first, a page at a time (next_cursor), with total and total_exact

get_market

One market's terms, status, result, and how much of it was recorded

market_price_history

The recorded YES bid/ask path, up to about 240 points over the market's busiest stretch; path_source is life, busiest or window

list_datasets

Every dataset: what it holds, columns, plan, days available

dataset_series

Which series have files in a dataset, with days, rows and size, up to 500 a page with total and next_offset

recording_status

Whether recording is running, and the recorded share of recent days by venue

list_files

Daily Parquet files for a dataset, date range and series, page by page

download_links

Signed download URLs for up to 100 of those files (counts toward the monthly allowance)

Things to ask:

  • "Which bitcoin markets does Kalshi Eye record, and since when?"

  • "What were the last few hours of the Bills vs Rams market on October 12 like?"

  • "How much of the Fed decision series is in the order book files?"

  • "Is the recorder running, and how complete was last week?"

Notes

  • Kalshi is recorded since October 12, 2026; Polymarket, Limitless and Chainlink prices since October 15, 2026. Each UTC day is published the next morning.

  • Prices are dollars between 0 and 1. A Kalshi YES ask is 1 minus the best NO bid. Times are UTC.

  • Free data without any key: a one-day sample on Hugging Face and runnable examples. The Python client for downloading files is kalshieye.

  • KALSHIEYE_BASE_URL points the server at another deployment (default https://kalshieye.com).

Kalshi Eye is independent and is not affiliated with Kalshi, Polymarket or Limitless. MIT licensed.

Available Tools

10 tools
dataset_seriesC
Read-only

Which series have files in one dataset (for example book_l2 or trades), with days, rows and size.

Inputs: dataset_id (required), limit, query. Returns: object

get_marketA
Read-only

One market by ticker (Kalshi ticker, or a Polymarket token id): terms, status, result, and how much of its life was recorded.

Inputs: ticker (required). Returns: object

get_seriesA
Read-only

One series (for example KXBTC15M or KXFEDDECISION): what it is, how often it runs, and how many markets and days the archive has.

Inputs: series_id (required). Returns: object

list_datasetsA
Read-only

Every dataset in the archive (order books, top of book, minute bars, trades, settlements, Polymarket, Chainlink...): what it holds, the plan it needs, days available, and its columns.

Inputs: no named parameters. Returns: object

list_filesB
Read-only

Daily Parquet files for a dataset and date range (YYYY-MM-DD, UTC), optionally one series. Needs KALSHIEYE_API_KEY.

Inputs: dataset (required), end_date (required), start_date (required), limit, +1 more. Returns: object

list_marketsA
Read-only

Markets in one series that closed or close within the last days days (1 to 7), with status, result and times.

status is a Kalshi market status, such as active or finalized.

Inputs: series (required), days, limit, status. Returns: object

market_price_historyA
Read-only

The recorded YES best bid and ask over a market's life, about 40 points, plus how much was recorded.

Good for "what did the market think over time". For every order book change, use the daily Parquet files (list_datasets, list_files).

Inputs: ticker (required). Returns: object

recording_statusA
Read-only

Whether the recorder is running now, the last published day, and the recorded share of recent days by venue.

Inputs: no named parameters. Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 10 tools

    Disambiguation4/5

    Each tool has a distinguishable target: search (keyword discovery), get_series/list_markets/get_market (drill-down by series ticker or market ticker), and a separate archive tier (list_datasets, dataset_series, list_files, download_links). The only mild overlaps are get_series vs dataset_series and search vs list_markets, but descriptions disambiguate them clearly.

    Naming Consistency3/5

    Most tools follow verb_noun (download_links, get_series, list_markets, get_market, list_datasets, list_files), but there are notable deviations: the bare verb 'search' and noun-only names like market_price_history, dataset_series, and recording_status. Still readable, but the pattern isn't uniform.

    Tool Count5/5

    Ten tools is well-scoped for a market data archive: discovery, metadata, file listing, and download each earn their place with no redundant surface.

    Completeness4/5

    The full pipeline is covered: search/discover markets and series, inspect datasets, list daily Parquet files, and obtain signed download URLs, plus a recorder health check. Minor gaps exist—events are only reachable via search, and there is no direct query tool beyond price history and raw file download.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables access to Polymarket's prediction markets for analyzing market probabilities, trading activity, and event outcomes across politics, sports, crypto, and other categories through natural language queries.
      10
      ISC
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI assistants to query Polymarket prediction markets, accessing real-time odds, market data, price history, order books, and trending markets across categories like politics, crypto, and sports through natural language.
      7
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      Provides search, trending, odds, arbitrage, and category browsing for prediction markets (Polymarket & Kalshi) via the Model Context Protocol, enabling AI agents to access live market data without API keys.
      5
      1
      MIT