npx skills add ...
npx skills add mapbox/mapbox-agent-skills --skill mapbox-geospatial-operations
npx skills add mapbox/mapbox-agent-skills --skill mapbox-geospatial-operations
Expert guidance on choosing the right geospatial tool based on problem type, accuracy requirements, and performance needs
Expert guidance for AI assistants on choosing the right geospatial tools from the Mapbox MCP Server. Focuses on selecting tools based on what the problem requires - geometric calculations vs routing, straight-line vs road network, and accuracy needs.
The Mapbox MCP Server provides two categories of geospatial tools:
The key question: What does the problem actually require?
| Problem Characteristic | Tool Category | Why |
|---|---|---|
| Straight-line distance (as the crow flies) | Offline geometric | Accurate for geometric distance |
| Road/path distance (as the crow drives) | Routing API | Only routing APIs know road networks |
| Travel time | Routing API | Requires routing with speed/traffic data |
| Point containment (is X inside Y?) | Offline geometric | Pure geometric operation |
| Geographic shapes (buffers, centroids, areas) | Offline geometric | Mathematical/geometric operations |
| Traffic-aware routing | Routing API | Requires real-time traffic data |
| Route optimization (best order to visit) | Routing API | Complex routing algorithm |
| High-frequency checks (e.g., real-time geofencing) | Offline geometric | Instant response, no latency |
User asks: "How far is X from Y?"
| What They Actually Mean | Tool Choice | Why |
|---|---|---|
| Straight-line distance (as the crow flies) | distance_tool | Accurate for geometric distance, instant |
| Driving distance (as the crow drives) | directions_tool | Only routing knows actual road distance |
| Walking/cycling distance (as the crow walks/bikes) | directions_tool | Need specific path network |
| Travel time | directions_tool or matrix_tool | Requires routing with speed data |
| Distance with current traffic | directions_tool (driving-traffic) | Need real-time traffic consideration |
Example: "What's the distance between these 5 warehouses?"
distance_tool (10 calculations, instant)matrix_tool (5×5 matrix, one API call, returns actual route distances)Key insight: Use the tool that matches what "distance" means in context. Always clarify: crow flies or crow drives?
User asks: "Which points are near/inside this area?"
| Query Type | Tool Choice | Why |
|---|---|---|
| "Within X meters radius" | distance_tool + filter | Simple geometric radius |
| "Within X minutes drive" | isochrone_tool → point_in_polygon_tool | Need routing for travel-time zone, then geometric containment |
| "Inside this polygon" | point_in_polygon_tool | Pure geometric containment test |
| "Reachable by car in 30 min" | isochrone_tool | Requires routing + traffic |
| "Nearest to this point" | distance_tool (geometric) or matrix_tool (routed) | Depends on definition of "nearest" |
Example: "Are these 200 addresses in our 30-minute delivery zone?"
isochrone_tool (routing API - need travel time)point_in_polygon_tool (geometric - 200 instant checks)Key insight: Routing for creating travel-time zones, geometric for containment checks
User asks: "What's the best route?"
| Scenario | Tool Choice | Why |
|---|---|---|
| A to B directions | directions_tool | Turn-by-turn routing |
| Optimal order for multiple stops | optimization_tool | Solves traveling salesman problem |
| Clean GPS trace | map_matching_tool | Snaps to road network |
| Just need bearing/compass direction | bearing_tool | Simple geometric calculation |
| Route with traffic | directions_tool (driving-traffic) | Real-time traffic awareness |
| Fixed-order waypoints | directions_tool with waypoints | Routing through specific points |
Example: "Navigate from hotel to airport"
directions_toolbearing_toolKey insight: Routing tools for actual navigation, geometric tools for directional info
User asks: "Create a zone around this location"
| Requirement | Tool Choice | Why |
|---|---|---|
| Simple circular buffer | buffer_tool | Geometric circle/radius |
| Travel-time zone | isochrone_tool | Based on routing network |
| Calculate area size | area_tool | Geometric calculation |
| Simplify complex boundary | simplify_tool | Geometric simplification |
| Find center of shape | centroid_tool | Geometric centroid |
Example: "Show 5km coverage around each store"
buffer_tool (geometric circles)isochrone_tool (routing-based)Key insight: Geometric tools for distance-based zones, routing tools for time-based zones
Small operations (< 100 calculations):
Medium operations (100-1,000 calculations):
Large operations (> 1,000 calculations):
Key insight: Volume rarely affects geometric tool choice, but routing APIs have batch tools for efficiency
| Use Case | Approach | Tool Choice |
|---|---|---|
| Real-time geofencing (every second) | Geometric checks | point_in_polygon_tool (instant) |
| Route planning (one-time) | Full routing | directions_tool or optimization_tool |
| Periodic proximity checks | Geometric distance | distance_tool |
| Live traffic routing | Routing with traffic | directions_tool (driving-traffic) |
Architecture note for high-frequency geofencing: If the application calls containment checks at very high rates (e.g., 50 vehicles every 2 seconds = 25 checks/second), calling MCP tools over the network adds agent-reasoning overhead that makes it impractical. In those cases, recommend using Turf.js directly in-process (turf.booleanPointInPolygon) for the hot path, and reserve MCP tools for peripheral tasks like zone definition (isochrone_tool), rerouting (directions_tool), or visualization (static_map_image_tool).
User: "Find the closest store and show 5km coverage"
Optimal approach:
category_search_tool (returns distances automatically)buffer_tool (5km geometric circle)static_map_image_toolWhy: Search already gives distances; geometric buffer for simple radius
User: "Optimize delivery to 8 addresses / stops"
Optimal approach:
search_and_geocode_tool to convert any street addresses to coordinates. Even when coordinates are already provided, mention this as an optional pre-step — real-world delivery lists often contain a mix of addresses and coordinates.optimization_tool (TSP solver — reorders stops to minimize total drive time)Why optimization_tool and NOT these alternatives:
directions_tool only routes A → B (or through fixed-order waypoints). It does NOT reorder stops — if you pass 8 stops, it routes them in the order given, which is almost never optimal.matrix_tool gives travel times between all pairs of stops (8×8 = 64 values), but it does NOT compute the optimal ordering. You'd need to solve TSP yourself on top of the matrix — optimization_tool does this for you in one call.Always mention search_and_geocode_tool as a useful companion for geocoding delivery addresses before optimization.
User: "Which of these 200 addresses can we deliver to in 30 minutes?"
Optimal approach:
isochrone_tool (30-minute driving)point_in_polygon_tool (200 geometric checks)Why: Routing for accurate travel-time zone, geometric for fast containment checks
User: "How long was this bike ride?"
Optimal approach:
map_matching_tool (snap to bike paths)distance_toolWhy: Need road/path matching; distance calculation either way works
User: "What's our total service area?"
Optimal approach:
buffer_toolarea_toolisochrone_tool for each locationWhy: Geometric for distance-based coverage, routing for time-based
Why wrong: As the crow flies ≠ as the crow drives
Why wrong: Routing APIs don't do geometric containment
Why wrong: Roads aren't straight lines; traffic varies
Why better: Simpler, instant, answers the actual question
Some problems benefit from using both geometric and routing tools:
Use case: "Find gas stations along my route"
Use case: Quickly narrow down, then get precise routing for finalists
Use case: "Which customers are in our delivery zone?"
When user asks a geospatial question:
Before choosing a tool, ask:
Does "distance" mean as the crow flies or as the crow drives?
Does the user need travel time?
Is this about roads/paths or pure spatial relationships?
Does this need to happen in real-time with low latency?
Is accuracy critical, or is approximation OK?
Understanding what users mean:
| User Says | Usually Means | Tool Type |
|---|---|---|
| "Distance" | Context-dependent! Ask: crow flies or crow drives? | Varies |
| "How far" | Often as the crow drives (road distance) | Routing API |
| "Nearby" | Usually as the crow flies (straight-line radius) | Geometric |
| "Close" | Could be either - clarify! | Ask |
| "Reachable" | Travel-time based (crow drives with traffic) | Routing API |
| "Inside/contains" | Geometric containment | Geometric |
| "Navigate/directions" | Turn-by-turn routing | Routing API |
| "Bearing/direction" | Compass direction (crow flies) | Geometric |
distance_tool - Straight-line distance between two pointsbearing_tool - Compass direction from A to Bmidpoint_tool - Midpoint between two pointspoint_in_polygon_tool - Is point inside polygon?area_tool - Calculate polygon areabuffer_tool - Create circular buffer/zonecentroid_tool - Geometric center of polygonbbox_tool - Min/max coordinates of geometrysimplify_tool - Reduce geometry complexitydirections_tool - Turn-by-turn routingmatrix_tool - Many-to-many travel timesoptimization_tool - Route optimization (TSP)isochrone_tool - Travel-time zonesmap_matching_tool - Snap GPS to roadsUse Geometric Tools When:
Use Routing APIs When:
Works with:
// WRONG: Check if point is in polygon
// (Can't do this with routing APIs)
// CORRECT: Pure geometric operation
point_in_polygon_tool({ point: location, polygon: boundary });// User asks: "What's reachable in 20 minutes?"
// WRONG: 20-minute distance at average speed
distance_tool + calculate 20min * avg_speed
// CORRECT: Actual routing with road network
isochrone_tool({
coordinates: {longitude: startLng, latitude: startLat},
contours_minutes: [20],
profile: "mapbox/driving"
})// User asks: "Which direction is the airport?"
// OVERCOMPLICATED: Full routing
directions_tool({
coordinates: [
{ longitude: hotel[0], latitude: hotel[1] },
{ longitude: airport[0], latitude: airport[1] }
]
});
// BETTER: Just need bearing
bearing_tool({ from: hotel, to: airport });
// Returns: "Northeast (45°)"1. directions_tool → Get route geometry
2. buffer_tool → Create corridor around route
3. category_search_tool → Find POIs in corridor
4. point_in_polygon_tool → Filter to those actually along route1. category_search_tool → Find 10 nearby locations
2. distance_tool → Calculate straight-line distances (geometric)
3. For top 3, use directions_tool → Get actual driving time1. isochrone_tool → Create travel-time zone (routing)
2. point_in_polygon_tool → Check hundreds of addresses (geometric)1. Does it require routing, roads, or travel times?
YES → Use routing API (directions, matrix, isochrone, optimization)
NO → Continue
2. Does it require traffic awareness?
YES → Use directions_tool or isochrone_tool with traffic profile
NO → Continue
3. Is it a geometric/spatial operation?
- Distance between points (straight-line) → distance_tool
- Point containment → point_in_polygon_tool
- Area calculation → area_tool
- Buffer/zone → buffer_tool
- Direction/bearing → bearing_tool
- Geometric center → centroid_tool
- Bounding box → bounding_box_tool
- Simplification → simplify_tool
4. Is it a search/discovery operation?
YES → Use search tools (search_and_geocode, category_search)