HN Showcase
Back to browse
Natural language search across Kalshi and Polymarket (API and MCP)

Natural language search across Kalshi and Polymarket (API and MCP)

by helloiamvu·Feb 21, 2026·2 points·0 comments
Visit ProjectView on HN

AI Analysis

●●●BangerSolve My ProblemSlick

Solves messy cross-platform prediction market data with Claude-powered natural language search.

Strengths
  • •Tackles genuine data normalization problem: Kalshi and Polymarket structure identical contracts completely differently
  • •Three-tier access (web UI, API, MCP) means traders and agents use same search backend
  • •Live with 80k contracts, real product today, not a proof-of-concept
Weaknesses
  • •Relies on Claude as search backbone; unclear if performance or cost scales to enterprise query volume
  • •No public pricing or SLA mentioned, limiting trust for serious traders
Category
SaaS
Target Audience

Prediction market traders, AI agents querying markets, finance automation

Similar To

Polymarket search tools · Kalshi native search

Post Description

Hi HN, we built a better search for Kalshi and Polymarket. You can use it on the web, via API, or through MCP so AI agents can query it too.

There are about 80k active contracts across both platforms right now. Sports, weather, crypto, politics. Tonight's NBA games, BTC up or down in 5 mins. You name it. We trade prediction markets everyday and needed a smarter way to search across all of them.

We started by ingesting everything into Postgres and writing SQL to get insights. Then we plugged in Claude to query the DB for us. Type a question, get structured results. That actually worked and felt like what search should be. But the answers were only as good as the data, and the data was a mess.

The core problem: both platforms structure their data completely differently. One NBA game on Kalshi is dozens of separate contracts, each with its own ticker. Polymarket has the same game as a handful of contracts with different naming. One says "Cleveland," the other says "Cavaliers." You can't search across any of it without cleaning it up first.

So we built a pipeline to clean the data and classify it properly. Every market goes through it. Structured parsing for the predictable stuff, LLM for free-form titles/descriptions the rules can't handle. New markets get picked up and classified within minutes. Not glamorous work but it's what makes the search return the right results.

Some things you can try:

"NBA tonight" — games from both platforms resolving today

"Zelensky markets on Polymarket" — filtered to one platform

"Weather in Chicago today" — Kalshi has an entire weather derivatives market

"Kalshi trending" — sorted by volume

"pokemon" — you'd be surprised what people bet on

If you want to use it programmatically, there's a REST API (GET /api/search?q=...) and an MCP server at attena-api.fly.dev/mcp so AI agents can search prediction markets as a native tool. Docs at attena.xyz/docs/search.

Not everything works perfectly yet. But we built this for ourselves and figured others might find it useful too.

Try it here: https://www.attena.xyz/

Similar Projects