agrisignal-mcp

An MCP server for farm and land decision support, built entirely on free, keyless public data: Open-Meteo (weather, soil moisture/temperature, evapotranspiration) and SoilGrids (ISRIC's global 250m-resolution soil map). No API keys, no signup, no OAuth — every tool works out of the box.

Unlike general-purpose weather/soil API wrappers, this server doesn't just hand back raw data — it synthesizes weather and soil into farming decisions: whether to irrigate and how much, frost and heat stress risk windows, growing degree day accumulation for predicting crop development, and drought tracking. The goal is to be useful to someone without access to expensive precision-agriculture services — a smallholder farmer, an agronomy student, an extension worker, an NGO.

Every recommendation shows its underlying numbers and is framed as a data-grounded estimate, not certified agronomic advice. These are simplified models (see each tool's caveats below) meant for triage and planning, not a substitute for site-specific expertise.

Tools

| Tool | Description | |---|---| | geocode_location | Resolve a place name to coordinates (required first step for the others) | | get_soil_profile | Soil texture, pH, organic carbon at multiple depths (SoilGrids) | | get_growing_conditions | Current weather + soil snapshot + short forecast | | get_irrigation_advice | Soil water balance estimate: should you irrigate, and how much? | | get_frost_and_heat_risk | Scans the forecast for frost/freeze and heat-stress risk days | | get_growing_degree_days | GDD accumulation over a date range (crop development/maturity tracking) | | get_dry_spell_status | Consecutive dry days and total precipitation — a rough drought indicator |

Setup

No API keys needed. Add it to your MCP client and go.

Claude Code:

claude mcp add --transport stdio agrisignal -- npx -y agrisignal-mcp

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "agrisignal": {
      "command": "npx",
      "args": ["-y", "agrisignal-mcp"]
    }
  }
}

Any other MCP client that supports stdio servers can run this the same way: npx -y agrisignal-mcp.

Example usage

• "Should I irrigate my field near Ames, Iowa this week?" → geocode_location then get_irrigation_advice
• "What's the soil like at this location — good for growing tomatoes?" → get_soil_profile
• "Is there a frost risk in the next 10 days?" → get_frost_and_heat_risk
• "How many growing degree days have accumulated since I planted on May 15?" → get_growing_degree_days
• "Has this region been in a dry spell?" → get_dry_spell_status

Notes on the models used

• Irrigation advice is a simplified single-layer soil water balance (in the spirit of FAO-56 accounting): total available water = (field capacity − wilting point) × root zone depth, projected forward using forecast precipitation minus reference evapotranspiration (ET0, not crop-specific actual ET). It ignores runoff and drainage below the root zone. Field capacity and wilting point come from SoilGrids at 15–30cm depth; current soil moisture from Open-Meteo's 9–27cm band — a reasonable but approximate match to a generic root zone, not a specific crop's.
• Growing degree days use the modified method common in US extension-service guidance for corn: Tmin is floored at the base temperature before averaging, and an optional Tmax cap can be applied. Set floor_tmin_at_base: false for the plain average method if that fits your crop better.
• Dry spell severity thresholds (mild >7 days, moderate >14, severe >21) are a general rule of thumb, not a calibrated meteorological drought index (e.g. SPI).
• SoilGrids no-data pixels: geocoded place coordinates often land on a town/city center, which can be a no-data pixel in SoilGrids' 250m grid (buildings, pavement). Soil-data tools automatically retry ~1km away and say so in the response when this happens — for a real field, pass its actual coordinates rather than a town center for a more accurate reading.

Rate limits

Open-Meteo and SoilGrids are free services. Be considerate — avoid hammering either API with rapid repeated calls; SoilGrids in particular can take several seconds per request.

Development

npm install
npm run build
npm test

License

MIT