EmbedEarth gives AI agents an open, programmable layer for the physical world: search places and infrastructure, run spatial analysis, route journeys, prepare MapLibre sources, and work with local geographic data.

The MCP server has the same six core areas as the SDK and CLI: Search, Compute, Routing, Maps, Catalog, and Stratum.

Product: embed.earth · Docs: MCP docs · Docs home

Run

npx -y @embed-earth/mcp-test

Configure any stdio-compatible MCP client:

{
  "mcpServers": {
    "e2": {
      "command": "npx",
      "args": ["-y", "@embed-earth/mcp-test"],
      "env": {
        "E2_API_KEY": "your_api_key"
      }
    }
  }
}

Enable Cloud

Create an API key at embed.earth and set it as E2_API_KEY in the MCP host environment. Cloud requests require the key, and Street View image URLs also require a Cloud API key. Do not commit real keys to an MCP configuration. Routing uses https://route.embed.earth by default; ROUTE_URL is only needed to override it.

Search

e2_search returns a GeoJSON FeatureCollection. Use area for a human-readable name or area_id for an exact raw catalog ID.

{
  "feature": "restaurant",
  "area": "New York",
  "limit": 25,
  "mode": "auto"
}

{
  "feature": ["restaurant", "hospital"],
  "area_id": "4915",
  "bbox": [-74.1, 40.65, -73.9, 40.85],
  "limit": 100
}

Input	Purpose
`feature`	Required feature key/ID or array of keys/IDs.
`area` / `area_id`	Named area or exact raw area ID.
`bbox`	Optional `[minLng, minLat, maxLng, maxLat]` boundary.
`where` / `properties`	Optional source-column or property filters.
`mode` / `limit`	Execution backend and maximum returned records.

Compute

e2_compute evaluates a spatial relation or aggregate. Use e2_compare for two features in one area or one feature across two areas.

Primitive	Purpose
`NEAREST`	Find the closest selected features to a relation feature.
`DISTANCE`	Measure distances between selected and relation features.
`WITHIN`	Return selected features inside a specified relation distance.
`COUNT`	Count selected features that meet the relation.
`DENSITY`	Measure feature concentration in an area.
`COVERAGE`	Measure area reached by a selected relation.
`GAPS`	Identify places not reached by selected coverage.
`CLUSTER`	Group selected features by spatial proximity.

{
  "primitive": "COUNT",
  "feature": ["restaurant"],
  "area_id": "4915",
  "near": "park",
  "distanceMeters": 500,
  "mode": "cloud"
}

e2_compute accepts feature, area or area_id, optional near or within, distanceMeters, limit, mode, and an optional local database.

Routing

e2_route runs named operations through https://route.embed.earth.

Operation	Purpose
`route`	Route between ordered locations.
`matrix`	Return travel metrics for source/target pairs.
`isochrone`	Return reachable contours.
`mapMatch`	Align a coordinate trace to the network.
`locate`	Return network information near locations.
`height`	Return elevation information for locations.
`e2_route_to_feature`	Find the nearest indexed destination, then route to it.

{
  "operation": "route",
  "body": {
    "locations": [{ "lat": 40.74, "lon": -73.99 }, { "lat": 40.71, "lon": -74.01 }],
    "mode": "walking"
  }
}

Maps

e2_map_sources returns MapLibre-ready source descriptors for the Planet basemap and selected published feature PMTiles. It does not create a map in the MCP host.

{ "features": ["restaurant", "hospital"] }

Catalog

e2_catalog is the single catalog lookup tool. It lists, searches, and inspects features, areas, and data; it does not expose a resolver.

{ "kind": "areas", "action": "search", "query": "New York" }

{ "kind": "areas", "action": "info", "id": "4915" }

Use raw area IDs such as 4915, without a provider prefix.

Stratum

e2_stratum is the local-data tool. It manages the verified cache, pulls data from the server, queries it in Cloud or local mode, downloads files, and uploads local GeoJSON, CSV, or Parquet.

`operation`	Required inputs	Purpose
`cacheList`	none	Inspect the local cache.
`cacheSet`	`size`	Set the cache limit (minimum 1GB).
`cacheDelete` / `cacheClear`	`key` for delete	Remove cached data.
`download`	`feature`, optional `area_id`, `output`	Pull verified data to a file or cache.
`query`	`feature`, optional `area_id`	Query server or verified local data.
`upload`	`path`, `database`	Index local GeoJSON, CSV, or Parquet.

{
  "operation": "upload",
  "path": "./places.geojson",
  "database": "./places.db",
  "connectTo": "restaurant"
}

Choose Search for records, Compute for spatial analysis, Routing for network travel, Maps for source descriptors, Catalog for valid inputs, and Stratum for server pulls and local data.