The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Landbenchmark MCP listing page.
An MCP server that lets AI agents run satellite land due-diligence through LandBenchmark. It exposes one tool, analyze_parcel, that returns a green / caution / walk-away verdict with cited signals (flooding, slope & buildability, soil, hazards, access) for any parcel.
When someone asks their AI assistant "is this land any good to buy?", the agent can call LandBenchmark and answer with observed, cited satellite data instead of guessing.
Both env vars are required. (TERRAIN_* is the internal engine name — LandBenchmark runs on the Terrain analysis engine.) TERRAIN_BASE_URL has no default on purpose: every request sends your API key to that host in an Authorization header, so the server refuses to start rather than guess where it goes. It also refuses to send a key over plain http:// to anything but localhost.
Works with any MCP-capable client — Claude Desktop, Claude Code, and agent frameworks.
analyze_parcel| Param | Type | Notes |
|---|---|---|
lat, lon | number | Parcel centre (WGS84). Provide these or geometry. |
geometry | GeoJSON | Polygon or Point (alternative to lat/lon). |
label | string | Optional parcel name. |
mode | "lite" | "full" | lite (default) ≈ 1 min; full = deep multi-year satellite report ≈ 3–4 min. |
Returns a plain-text verdict summary with the flagged signals and a link to the full report.
Informational only — not a survey, flood determination, or a substitute for on-site inspection and professional advice.