grounding-lite

name	grounding-lite
description	Google Maps Grounding Lite MCP for location search, weather, and routes via mcporter.
homepage	https://developers.google.com/maps/ai/grounding-lite
metadata	{"clawdbot":{"emoji":"🗺️","requires":{"bins":["mcporter"],"env":["GOOGLE_MAPS_API_KEY"]},"primaryEnv":"GOOGLE_MAPS_API_KEY","install":[{"id":"node","kind":"node","package":"mcporter","bins":["mcporter"],"label":"Install mcporter (npm)"}]}}

Grounding Lite

Google Maps Grounding Lite MCP provides AI-grounded location data. Experimental (pre-GA), free during preview.

Setup

1. Enable the API: gcloud beta services enable mapstools.googleapis.com
2. Get an API key from Cloud Console
3. Set env: export GOOGLE_MAPS_API_KEY="YOUR_KEY"
4. Configure mcporter:

   mcporter config add grounding-lite \
     --url https://mapstools.googleapis.com/mcp \
     --header "X-Goog-Api-Key=$GOOGLE_MAPS_API_KEY" \
     --system
   

Tools

• search_places: Find places, businesses, addresses. Returns AI summaries with Google Maps links.
• lookup_weather: Current conditions and forecasts (hourly 48h, daily 7 days).
• compute_routes: Travel distance and duration (no turn-by-turn directions).

Commands

# Search places
mcporter call grounding-lite.search_places textQuery="pizza near Times Square NYC"

# Weather
mcporter call grounding-lite.lookup_weather location='{"address":"San Francisco, CA"}' unitsSystem=IMPERIAL

# Routes
mcporter call grounding-lite.compute_routes origin='{"address":"SF"}' destination='{"address":"LA"}' travelMode=DRIVE

# List tools
mcporter list grounding-lite --schema

Parameters

search_places: textQuery (required), locationBias, languageCode, regionCode

lookup_weather: location (required: address/latLng/placeId), unitsSystem, date, hour

compute_routes: origin, destination (required), travelMode (DRIVE/WALK)

Notes

• Rate limits: search_places 100 QPM (1k/day), lookup_weather 300 QPM, compute_routes 300 QPM
• Include Google Maps links in user-facing output (attribution required)
• Only use with models that don't train on input data