Microfarm MCP

Land-viability intelligence for would-be small farmers — a free, open-source MCP server that plugs into Claude, Cursor, or any MCP-capable AI to answer the two questions that matter about a piece of land:

1. Is this plot actually good for growing food?

2. Can I legally put a home on it?

…and finds real land for sale to score in the first place.

A tool for stewards, not speculators.

Why this exists

Everyone should own land — because this is what will make you rich. Not money, not nice cars — but land you can cultivate, grow, and enrich; land that provides for you and your family, your neighbours, and your community; land that feeds animals and sustains life.

— the reason this project exists

Most of what the real-estate market does with land is speculating on it. This tool is for the people who want a piece of ground to own, live on, and care for — water, topsoil, a workable season, and the honest answer about whether they're actually allowed to build on it. It does not guess: every signal comes from a named public source, and anything it can't know is flagged, never invented.

Works in the US, France, and Cyprus — and scores land anywhere Open-Meteo + OSM have data. Exactly what's available differs by country, so here's the honest picture:

Related MCP server: Plantos MCP Server

Country coverage

Verdict legend: ✅ = works now · ⚠️ = works, with an honest "confirm locally" note · 🔜 = registered but needs a browser (see land_sources.browser_only)

What's global vs. country-specific:

• Global (works almost anywhere): climate (precipitation, growing season, hardiness), water features, farmland, land type, roads/access — this is what assess_land scores, and it uses Open-Meteo + OpenStreetMap, both worldwide.

• Country-specific: find_land listings come from per-country sites; flood zones come from FEMA, which is US-only — so the dwelling-flood check really only applies to the US, and elsewhere it honestly says "confirm with your local planning authority" rather than guessing.

Tools

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

Quickstart

git clone https://github.com/omardaaboul/microfarm-mcp.git
cd microfarm-mcp
python3 -m venv .venv
.venv/bin/pip install -e ".[test]"
.venv/bin/microfarm-mcp            # start the server (stdio)

Connect it to Claude / Cursor / OpenCode / Gemini / any MCP client

Microfarm is a standard MCP server, so any tool that supports MCP can use it. The exact setup depends on the client:

Claude Desktop — claude_desktop_config.json:

{
  "mcpServers": {
    "microfarm-mcp": {
      "command": "python3",
      "args": ["-m", "microfarm.server"],
      "env": { "PYTHONPATH": "/absolute/path/to/microfarm-mcp" }
    }
  }
}

Claude Code — claude mcp add:

claude mcp add microfarm-mcp -- python3 -m microfarm.server

Cursor — .cursor/mcp.json:

{
  "mcpServers": {
    "microfarm-mcp": {
      "command": "python3",
      "args": ["-m", "microfarm.server"],
      "env": { "PYTHONPATH": "/absolute/path/to/microfarm-mcp" }
    }
  }
}

OpenCode — opencode.json (repo) or ~/.config/opencode/config.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "microfarm-mcp": {
      "type": "local",
      "command": ["python3", "-m", "microfarm.server"],
      "cwd": "/absolute/path/to/microfarm-mcp",
      "enabled": true,
      "timeout": 10000
    }
  }
}

Gemini CLI — ~/.gemini/settings.json:

{
  "mcpServers": {
    "microfarm-mcp": {
      "command": ["python3", "-m", "microfarm.server"],
      "env": { "PYTHONPATH": "/absolute/path/to/microfarm-mcp" }
    }
  }
}

VS Code (built-in Copilot MCP / extension) — .vscode/mcp.json:

{
  "servers": {
    "microfarm-mcp": {
      "type": "stdio",
      "command": "python3",
      "args": ["-m", "microfarm.server"],
      "env": { "PYTHONPATH": "/absolute/path/to/microfarm-mcp" }
    }
  }
}

A note on ChatGPT. ChatGPT supports MCP only as a remote HTTPS server, and that capability is currently limited to Business/Enterprise plans. A local stdio server (what this repo runs by default) cannot be plugged into ChatGPT directly. To use it in ChatGPT you'd need to host the server remotely and give ChatGPT its HTTPS URL — a real step, not a config tweak. For local use, Claude, Cursor, OpenCode, or Gemini CLI are the simple paths.

Then just ask your assistant:

"Is this plot in Dordogne good for a small farm, and can I build a house on it?"

"Find me land under €150k in Paphos District and tell me which looks worth a visit."

What the results look like

assess_land("Polemi, Paphos, Cyprus")
→ promising | 75
  [positive] adequate_precipitation: ~426 mm/yr
  [positive] long_season:            ~363 growing days/yr
  [positive] water_access:           5 water feature(s) nearby
  [positive] farmland:               27 farm land tag(s) nearby
  [caution]  land_type:              built-up
  [caution]  tree_cover:             little/no mapped tree cover
  [caution]  privacy:                23 structures nearby
  [positive] soil_fertility:         ~42 g/kg organic carbon
  [caution]  soil_ph:                pH ~7.8
  [caution]  well_recharge:          ~33% clay / ~23% sand
  [positive] access:                 158 road segment(s) nearby

zone("A", "France")   → explains agricultural zone + when a dwelling is allowed
zone("Γη3", "Cyprus") → agricultural zone, min ~4,000m² for one rural dwelling

Data sources & attribution

These are the free, keyless public sources behind every answer. When you use this project or its data in your own work, please credit the sources below — several require attribution by license.

Open-Meteo aggregates national weather services; its API is licensed CC BY 4.0 and we gratefully use it. OpenStreetMap data is © OpenStreetMap contributors under the Open Database License. Listing content belongs to the respective marketplaces — we surface and link through to their pages, never re-host their images.

Land type, zone & the "can I build a dwelling here?" question

• Land type (farmland / pasture / forest / scrub / built-up / waterfront) and tree cover are detected from OpenStreetMap and reported in every assessment.

• Official zoning — what you're legally allowed to build or do on land — is not available from free national data. It's set by local planning authorities (US county, Cyprus Planning Authority, French mairie/DDT), and recorded on title deeds / cadastres / PLU documents that aren't freely machine-queryable. The zone tool explains the codes (France A/U/N, Cyprus Γη/ΚΑ, US agricultural districts) and their usual building rules — including the specific conditions for a farm dwelling / logement de fonction — and tells you exactly what to confirm with the local authority. It is a guide, never a substitute for the official record.

Honest limits

• Zoning / dwelling law is local. This tool is an initial screen from public data (flood + farmland context); it does not replace the official ordinance. Always verify in person and with local officials.

• Flood data is US-only (FEMA). Outside the US it's honestly reported as unavailable — never falsely "no hazard."

• Some listing sites are browser-locked (landbanksearch, govdeals, farmflip, Bazaraki, land.cy, offer.com.cy, leboncoin.fr). They're registered but not yet scraped; land_sources shows the boundary of what's reachable vs. browser-only.

• Geo-region names like "Limousin" (abolished in 2016) may not geocode — use a current town or district.

• Greek listing titles get a basic word-level translation; obscure place names may stay in Greek.

Development

.venv/bin/ruff check microfarm tests   # lint
.venv/bin/ruff format --check microfarm tests
.venv/bin/mypy microfarm               # type-check
.venv/bin/pytest tests/ -q             # 43 tests (mocked providers, deterministic)

CI runs lint, format, type-check, tests, and a package build on Python 3.10 / 3.11 / 3.12.

Project layout

microfarm-mcp/
├── microfarm/
│   ├── server.py       ← FastMCP server (5 tools)
│   ├── engine.py       ← orchestration + cache + find_land
│   ├── providers.py    ← free data providers (Open-Meteo, SoilGrids, OSM, FEMA)
│   ├── listings.py     ← land-listing collectors (farms.com, Craigslist, index.cy)
│   ├── scorer.py       ← evidence-backed viability + dwelling scoring
│   ├── zones.py        ← planning-zone explainer (France, Cyprus, US)
│   └── models.py       ← typed pydantic models
├── tests/              ← 43 tests
├── configs/            ← ready-to-use MCP client configs (Claude, Cursor,
│                         OpenCode, Gemini CLI, VS Code, generic)
├── .github/workflows/ci.yml
└── pyproject.toml

License & disclaimer

MIT licensed — you're free to use, modify, and share it.

This is not financial, legal, or real-estate advice. Always verify a property in person and with the local planning authority before relying on it. Data comes from the free public sources above and may be incomplete or imperfect.