Microfarm MCP

Land-viability intelligence for would-be micro-farmers — as an MCP server that plugs into any LLM (Claude, Cursor, etc.). It answers two questions about a US parcel:

1. "Is this land actually good for a micro-farm?" — scores climate, water access, farmland context, access, and flood risk from free public data (no keys, no subscriptions).

2. "Can I build a house on it?" — a dwelling-feasibility screen using flood zone + farmland context, honest that final approval is set by your county.

It can also find land for sale from free sources and score the results.

Tools

Everything is evidence-backed: every signal names its source, and missing data is reported as missing — never treated as "good."

Data sources (all free, keyless)

• Open-Meteo — geocoding, historical climate (precipitation, temps, growing days), elevation

• OpenStreetMap / Overpass — nearby water, buildings, roads, farmland

• FEMA National Flood Hazard Layer — flood zone at a point (dwelling feasibility)

• farms.com + Craigslist + index.cy (Cyprus) — land-for-sale listings (find_land)

Install & run

python3 -m venv .venv
.venv/bin/pip install -e ".[test]"
.venv/bin/pytest tests/ -q        # 31 tests
.venv/bin/microfarm-mcp           # or: python -m microfarm.server

Example (what an LLM does for you)

assess_land("Petaluma, California")
→ state: promising | score: 76
  [positive] ample_precipitation: ~1023 mm/yr
  [positive] long_season: ~363 growing days/yr
  [positive] land_type: farmland
  dwelling_feasibility: requires_review (no flood-zone flag; confirm with county)

assess_land("Polemi, Paphos, Cyprus")   ← works outside the US too
→ state: requires_look | score: 65
  [positive] adequate_precipitation: ~457 mm/yr
  [positive] long_season: ~365 growing days/yr
  dwelling_feasibility: unknown (FEMA is US-only; confirm with Cyprus Planning Authority)

assess_land("Nontron, Dordogne, France")   ← France works too
→ state: promising | score: 76
  [positive] ample_precipitation: ~1416 mm/yr
  [positive] long_season: ~330 growing days/yr
  dwelling_feasibility: unknown (confirm with the local mairie / DDT planning office)

find_land(state="OH", max_price_usd=200000)
→ real listings with price, acres, location, photos, description

find_land(country="CY")   ← Cyprus plots from index.cy
→ real Greek/English listings, titles translated, price in €, size in m²/acres

Land type + zone

• Land type (farmland / pasture / forest / scrub / built-up / water) is detected from OpenStreetMap and reported in every assessment.

• Official zoning (what you're legally allowed to build/use the land for) is not available from free national data — it's set by local planning authorities (a US county, or a Cyprus District/Planning office). The tool flags this honestly and tells you where to confirm rather than guessing.

Honest limits

• Zoning/dwelling law is local; this uses flood + farmland context as an initial screen and points you to the planning office — it does not replace the official ordinance.

• FEMA flood data is US-only; outside the US (e.g. Cyprus) flood is honestly reported as unavailable, never falsely "no hazard."

• landbanksearch / govdeals / farmflip / Bazaraki / land.cy / offer.com.cy are browser-only or bot-protected; registered but not yet scraped (see land_sources.browser_only).

Layout

microfarm-mcp/
├── microfarm/
│   ├── server.py       ← FastMCP server (4 tools)
│   ├── engine.py       ← orchestration + cache + find_land
│   ├── providers.py    ← free data providers (geo, climate, elevation, OSM, FEMA)
│   ├── listings.py     ← land-for-sale collectors (farms.com, Craigslist)
│   ├── scorer.py       ← evidence-backed viability + dwelling scoring
│   └── models.py       ← typed models
├── tests/              ← 24 tests (mocked providers, deterministic)
├── .github/workflows/ci.yml
└── pyproject.toml

MIT licensed. Not financial, legal, or real-estate advice — verify every property in person.