Two years ago we started pricing the Kalshi daily-high temperature markets against National Weather Service forecasts. It became one of our most reliable reads — a public, objective resolution source, a forecast that's genuinely predictive, and a market that doesn't always agree with either. Today we did something we'd been building toward: we opened the whole edge feed to machines.
What we shipped
The signals we already generate — weather, the commodity ladders (bitcoin, silver, gold, oil), and daily mispricings — are now a single JSON feed:
- A public page,
/alerts— the feed, 24 hours delayed, free. Grouped by day, linked to the tool behind each signal, with our honest graded record on display. - A real-time API,
/api/v1/alerts— the same feed, live, for Pro keys. JSON envelope, keyset pagination, per-key rate limits. The samepmp_v2_pro_…key you already have. - An MCP tool —
edge_alertson our MCP server, so an AI agent can read the feed natively in Claude, ChatGPT, or Cursor. - An open-source, non-custodial executor — a script you run yourself, with your own Kalshi key, that turns alerts into orders under rules you set. It never sees our servers touch your keys, because they never do.
Why "to robots"
Because the natural consumer of a structured signal feed isn't a person refreshing a page — it's a script, or increasingly, an agent. An LLM with the MCP tool can ask "any STRONG weather edge on Kalshi right now?" and get a clean answer. A Python loop can poll every 60 seconds and act. We'd rather meet that where it lives than pretend everyone wants a dashboard.
The posture, on purpose
One thing we won't do: tailor the signal. Every subscriber — human or agent — gets the identical feed at the identical time. Filters let you select what you see (only weather, only STRONG, since a timestamp); they never change the signal itself. We publish; we don't advise, and we don't execute. The open-source executor is non-custodial precisely so that line stays clean: you hold the keys, you place the orders, you own the decisions.
The honest caveats
The weather edge is real, but two things temper it. Self-crowding — everyone acts on the same print, and Kalshi's weather books aren't infinitely deep, so the price can move before you fill (the executor ships with a price guard for exactly this). And fees eat small edges. We surface the graded record on /alerts — winners and losers — rather than a highlight reel, because the only honest way to judge a feed is over a lot of settled markets, net of what it costs to trade them.
Start here
Watch the free delayed feed. Read the API docs. Point an agent at the MCP server. And if you want to automate, run the executor in dry-run first — it logs what it would do and places nothing until you explicitly arm it.
Not investment advice. Impersonal analysis, published identically to everyone, at the same time. Trade responsibly.