Weather prediction markets
Live Kalshi contracts on daily high temperatures across US cities and on Eastern and Central Pacific storm counts. Each price is already a probability. These are the markets where the settlement rule matters most — the named station and the strictly-greater-than comparison decide more positions than the forecast does.
The same desk publishes this board as a remote MCP server, so an AI agent can price it without you opening a tab. Connect it to Claude, ChatGPT, Grok or Cursor — no API key, no account.
Quick Answer
Kalshi weather markets are daily high-temperature and Pacific storm-count contracts that settle at $1 or $0, so every price reads straight as a probability — and the named station plus the strictly-greater-than comparison decide more of them than the forecast does. There is also a weather MCP server: connect com.predictionmarketspicks/weather at https://predictionmarketspicks.com/api/mcp-weather/mcp and your agent can pull our live Kalshi weather edges, convert a price to an implied probability, test it against the climatological base rate, chain each new forecast run onto that prior, check expected value after fees, and return a Kelly-sized position. Free, read-only, no API key and no account.
Add it to your agent →Our record on weather contracts
−$7.93
net
−3.0¢
per contract
266
graded
35.0%
win rate
Net is a SUM over a flat one-contract stake; “per contract” is the average. Full method on the track record.
Add weather markets to your agent
One remote MCP server (Streamable HTTP transport), published on the official Model Context Protocol registry as com.predictionmarketspicks/weather under our DNS-verified namespace. No API key and no account needed to start calling.
https://predictionmarketspicks.com/api/mcp-weather/mcpclaude mcp add --transport http predictionmarketspicks-weather https://predictionmarketspicks.com/api/mcp-weather/mcp
Add this URL as a custom connector / remote MCP server: https://predictionmarketspicks.com/api/mcp-weather/mcp
Go to grok.com/connectors → New Connector → Custom, and paste the same URL: https://predictionmarketspicks.com/api/mcp-weather/mcp. No key, same as the others.
{
"mcpServers": {
"predictionmarketspicks-weather": {
"type": "http",
"url": "https://predictionmarketspicks.com/api/mcp-weather/mcp"
}
}
}VS Code: the same entry goes under "servers" (not "mcpServers") in .vscode/mcp.json.
There are two ways to connect, and they run the same weather desk on the same unlock:
- Weather only —
https://predictionmarketspicks.com/api/mcp-weather/mcp. Just the tools below, for a tidy tool list in your agent. Registry namecom.predictionmarketspicks/weather. - Full server —
https://predictionmarketspicks.com/api/mcp/mcp. Every tool below plus our prediction-market, commodity, NFL model and fantasy draft tools. Registry namecom.predictionmarketspicks/quant— see the full MCP server page.
Same caps, same unlock, same numbers either way. The full server is a superset — nothing here is weather-only-exclusive.
What your agent can call
One complete loop and no sports or fantasy noise: the signal, then the price read as a probability, then that price against the climatological base rate, then each new forecast run chained onto the prior, then whether the disagreement survives fees, then the size. Each tool returns structured numbers plus a link back to the same tool on the site.
Everything, at full depth, with one difference. Five of these are free calculators with no cap at all. edge_alerts is hybrid rather than gated: without a key the same live feed comes back delayed 24 hours with the written thesis stripped and a one-line offer attached, and a Pro key returns it in real time with the thesis. Nothing on this server is behind a Pro wall, and a keyless caller can never hit a hard refusal here — that is a property of the tool set, pinned by a test, not a claim in marketing copy.
The weather edges our models generate on Kalshi — temperature and precipitation contracts — as a live feed with side, price, model probability and edge. Pro keys get it in real time; without a key the same feed arrives delayed 24h.
Ask: “What is the edge on tomorrow’s NYC high?”
Turn a Kalshi weather contract price into an implied probability (and back) before comparing it to a forecast.
Ask: “Kalshi has Denver above 84 at 38c — what implied probability is that?”
Compare a weather contract price to the climatological base rate for that city and date window, with the gap in points and a sample-size quality read.
Ask: “Is that price off the climatological base rate for Denver in September?”
Update a temperature or precipitation probability as each new forecast run lands, chaining the evidence onto your prior step by step.
Ask: “The new forecast run moved the high up two degrees — update my probability.”
Expected value on a weather contract from its price and your forecast probability, with a BUY / SELL / SKIP signal and the fee drag named.
Ask: “Is that disagreement worth anything after fees?”
Size the position once the edge is real — full / half / quarter / eighth Kelly against your bankroll, with a risk rating.
Ask: “Size a position on the Denver ladder at quarter Kelly.”
Ask it, word for word
You never name a tool. Ask in plain language and the agent picks the call — these are the questions the loop above was assembled to answer, in the order a trader asks them.
- 1“What is the edge on tomorrow’s NYC high?”calls
edge_alerts - 2“Kalshi has Denver above 84 at 38c — what implied probability is that?”calls
convert_probability - 3“Is that price off the climatological base rate for Denver in September?”calls
base_rate_gap - 4“The new forecast run moved the high up two degrees — update my probability.”calls
bayes_update - 5“Is that disagreement worth anything after fees?”calls
calculate_ev - 6“Size a position on the Denver ladder at quarter Kelly.”calls
kelly_size
Read-only, all of it: the tools price a contract, they never place one, move money or write anything. You take the position on the exchange yourself. The same desk grades itself in public — every signal these engines publish is settled against the market that priced it, wins and losses both, at /track-record.
Open now
A contract price in cents is already a probability — 42c is a 42% implied chance. Prices from Kalshi, refreshed periodically; check the exchange before taking a position.
Price it yourself
Weather Edge →
Our temperature model against the Kalshi ladder, city by city — the engine behind the record above, and the same numbers the agent tools return.
How these contracts settle
Which station, which provider, and whether the comparison is strict — one page per series. On this board those three answers decide more positions than the forecast does.
Reading
Weather edges the morning the ladders open — temperature and storm counts. Free.
No spam. Unsubscribe anytime.
Weather markets and the weather MCP server — FAQ
Can you trade the weather on Kalshi?
Yes. Kalshi lists daily high-temperature contracts for a set of US cities and season-long contracts on Eastern and Central Pacific storm counts. Each trades between 1c and 99c on a CFTC-regulated exchange and settles at $1 or $0, so the price is the implied probability. It is a contracts exchange, not a sportsbook.
Which weather station decides a temperature contract?
One named station, and this is the single most expensive trap in the category. A city has several official sensors — an airport, a downtown climate station — and they routinely disagree by a few degrees on a hot afternoon. Only the station named in the rules counts, and only the named data provider’s figure counts, even if another source published a higher reading. The settlement page for each city names both.
Does hitting the threshold exactly resolve YES?
No. The comparison is strictly greater than, so a contract asking for more than 92 degrees resolves NO at exactly 92. On a contract that spends the whole afternoon within a degree of its strike, that one word is the entire position.
Why is the win rate under 50% when the record is positive?
Because the model is not trying to be right often — it is trying to buy contracts priced below what they are worth. Those tend to be cheap tail outcomes that pay several times the stake, so most positions lose a little and the winners carry the book. Judge a strategy like that on net result per contract, never on win rate; the two point in opposite directions by design.
Are there Atlantic hurricane contracts on Kalshi?
Not right now. Every open contract in the hurricane and named-storm series is Eastern Pacific or Central Pacific — checked against the exchange, with no Atlantic basin market in either series and no separate Atlantic series listed. That is worth knowing before you go looking: a storm-count position taken here is a Pacific position, and Atlantic activity does not settle it.
How far ahead are weather contracts listed?
Daily temperature ladders open the morning of, or the day before, and settle that evening, so the whole cycle is measured in hours. Storm-count contracts run the length of the season and settle at the end of it, which makes them the one weather market where a position can be held for months.
Is there a weather MCP server for AI agents?
Yes. Connect com.predictionmarketspicks/weather at https://predictionmarketspicks.com/api/mcp-weather/mcp (Streamable HTTP) and your agent gets 6 tools and nothing else: the live edge feed plus the calculators that turn a forecast into a position — edge_alerts, convert_probability, base_rate_gap, bayes_update, calculate_ev, kelly_size. No API key, no email, no account. The full server at https://predictionmarketspicks.com/api/mcp/mcp carries these same tools alongside our prediction-market and NFL model tools, and both work on the same unlock with the same depth limits. Pick the weather-only server if you want a tidy tool list in your agent; pick the full server if you want everything.
How do I add weather markets to Claude, ChatGPT or Grok?
In Claude Code, run: claude mcp add --transport http predictionmarketspicks-weather https://predictionmarketspicks.com/api/mcp-weather/mcp. In Claude Desktop, Claude.ai, ChatGPT or Cursor, add that same URL as a custom connector / remote MCP server. In Grok, go to grok.com/connectors, choose New Connector then Custom, and paste the same URL. Once connected, ask your agent things like "what is the edge on tomorrow's NYC high" or "size a position on the Denver ladder" and it calls the weather tools directly. Nothing to install and no account to create.
Can Claude trade weather markets?
It can price them, not place them. Every tool on the weather server is read-only: the agent can pull our live Kalshi weather edges, turn a contract price into an implied probability, compare it to the climatological base rate, update that probability as each forecast run lands, check whether the disagreement survives fees, and return a Kelly-sized position. It never routes an order, never moves money and never writes anything. You take the position yourself on the exchange.
What does the weather MCP server cost?
Nothing. Five of the 6 tools are free calculators with no cap at all. The sixth, edge_alerts, is hybrid rather than gated: without a key the same live feed arrives delayed 24 hours with the written thesis stripped, and a Pro key ($14.99/mo) returns it in real time with the thesis attached. A keyless caller can never hit a hard refusal on this server — that is a property of the tool set, locked by a test, not a promise in marketing copy.
Do I need an API key for the weather MCP server?
No. There is no key to claim, no email to hand over and no account to create — connect the URL and call. A key only changes one thing: it moves the edge feed from 24-hour delayed to real time and attaches the model thesis behind each signal. Everything else answers identically with or without one, and rate limiting is per IP.
Which tools does the weather MCP server expose?
edge_alerts, convert_probability, base_rate_gap, bayes_update, calculate_ev, kelly_size — one complete loop and no sports or fantasy noise. edge_alerts is the signal: our live Kalshi weather edges with side, price, model probability and edge. convert_probability turns a contract price into an implied probability. base_rate_gap compares that price to the climatological base rate for the city and date window. bayes_update chains each new forecast run onto that prior. calculate_ev checks whether the disagreement is worth anything after fees. kelly_size sizes the position. Every one of them is the same object the full server registers, so the numbers cannot disagree between the two.