Ask any weather question in plain language. Get the forecast fused with 35 years of ERA5-Land historical context — so you know not just what's coming, but whether it's normal, unusual, or worth acting on.
No AI-generated numbers. Every figure above comes from ERA5-Land reanalysis or live forecast data — a language model interprets your question, it never invents an answer.
Example figures — forecast & water-year confirmed live Aug 6, 2026; ski comparison confirmed live Aug 18, 2026.
The examples above are real, but they're fixed. This is the live app — ask about your own location and get the same kind of answer.
Opens the full app in place. For heavier use, see API access below.
A Telegram bot for quick questions and saved locations. Currently in development — not accepting new users yet.
The full experience — deeper comparisons, mapped locations, and the same app embedded above. Free, no key required, capped at 50 questions/day per visitor.
Programmatic access for agents, developers, and integrations. Structured, statistically rigorous, at volume. Free tier: 100 queries/day.
Baseline publishes an MCP server so agents and AI tools can call it directly — no scraping, no prompting a model to guess at climate numbers. Every tool call returns structured, sourced data.
Ask a weather or climate question in plain language. Returns forecast plus full historical context — percentile rank and anomaly against the 35-year record.
Same as above, for an exact latitude/longitude instead of a place name — for agents working from coordinates directly.
Precipitation and temperature status since the start of the water year (Oct 1 US / Jan 1 elsewhere), ranked against 35 years.
How unusual current or forecast conditions are at a single location — the "does this matter" question, answered directly.
Rank 2–10 locations against each other by precipitation, temperature, or snowfall — or use a curated category like colorado_ski_resorts.
pip install baseline-mcp · endpoint api.baselinecontext.com — also listed on the MCP registry (Glama, PulseMCP)
Get an API key →
You'll get a working key immediately — no approval wait, no card required. Early doesn't mean loose: when a question falls outside what the 35-year record actually supports, Baseline declines rather than guessing — the same discipline behind every number it does return. That's what makes an answer worth citing to someone else, not just worth reading yourself.
If you hit a gap, or a decline that shouldn't be one, tell us — that's exactly the edge being mapped right now. You're not just a user, you're helping define where "answered" and "declined" actually belong.