agribrain <!-- placeholder name — replace everywhere after Step 0 -->

Give your AI assistant a licensed agronomist's brain.

[DEMO GIF GOES HERE — 15s: real field coordinates → "What's the spray situation and water balance for my olives this week?" → real answer with numbers]

Every LLM can write a poem about olive trees. None of them know that your olive fruit fly third generation started Tuesday, that Saturday's wind makes spraying pointless, or that your field is 12mm behind on water. This MCP server fixes that — using only free, open data. No API keys. No accounts.

npx agribrain

Claude Desktop setup (60 seconds)

{
  "mcpServers": {
    "agri": { "command": "npx", "args": ["agribrain"] }
  }
}

Then ask: "Check the field briefing for 38.01, 23.72 — olives, planted March 2024."

Tools

Tool	What it answers
get_field_briefing	"What should I worry about this week?" — the everything report
get_spray_windows	"When can I actually spray?" — ranked windows with reasons
get_water_balance	"Am I irrigating enough?" — ET₀ loss vs. water received, net deficit
compute_gdd	"Where is the pest pressure?" — degree-days, generations, projected dates
get_chill_hours	"Did my orchard get enough winter chill?"
get_agro_weather	Forecast + recent history, in farming terms
get_soil_profile	pH, texture, organic carbon for any point on Earth

What makes this different

This is not another weather wrapper. Every model in this repo — pest degree-day thresholds, spray-window rules, crop staging — is curated and signed off by a licensed agronomist with 19 years in Mediterranean agriculture, with literature citations in the data files. Data is cheap; agronomy is the hard part.

• Decisions, not just data — spray windows, water deficits, generation timing
• Zero keys, zero cost — Met.no, NASA POWER, ISRIC SoilGrids (free, commercial-friendly)
• Citations included — every pest model links its sources
• Eval suite in the repo — we test that LLMs actually answer correctly with these tools

Data sources & attribution

Weather forecasts: MET Norway (CC-BY 4.0). Historical climate: NASA POWER (public domain). Soil: ISRIC SoilGrids (CC-BY 4.0). ET₀: computed via Hargreaves; FAO-56 (Allen et al., 1998).

Honest limitations

• GDD outputs without local trap data are estimates and labeled as such.
• SoilGrids is 250m resolution — a default, not a substitute for a soil test.
• This tool informs decisions; it does not replace your local agronomist or the product label. Nothing here is application-rate advice.

Roadmap

• v2 — eyes on the field: Sentinel-2 NDVI time series and zone anomaly detection for any field polygon (free Copernicus data), FAO-56 crop water demand (Kc × ET₀), 30-year climate context.
• v3 — compliance: EU pesticide approval checks, pre-harvest intervals, resistance groups.

Who builds this

Built and maintained as the open data layer of Ask Oli — the AI agronomist for Mediterranean smallholders. Maintained part-time by a solo founder; issues are triaged weekly, agronomy contributions (pest models for your region — see data/pest-models.json schema) are especially welcome and reviewed personally.

MIT licensed.