MA
OfficialSearchLocation

Mapbox MCP Server: Geocoding, POI Search, Directions, Isochrones

Unlock geospatial intelligence with the MCP server using Mapbox geocoding, POI search, directions and isochrones to power fast, scalable location services.

View on GitHub
Quick Install
npx -y @mapbox/mcp-server

Overview

The Mapbox MCP Server is a Node.js implementation of the Model Context Protocol (MCP) that exposes Mapbox location services as tools for AI agents and other MCP-capable clients. It acts as a bridge between general-purpose agents (chatbots, assistants, IDE integrations) and Mapbox geospatial APIs, enabling natural-language-driven access to geocoding, POI search, routing, isochrones, map images, and other spatial operations.

Running the server locally or using the hosted endpoint lets your application “understand where” — resolve addresses, find nearby businesses, generate travel-time areas, compute optimized routes, and render static maps — without each client having to implement many Mapbox calls. The MCP server handles request translation, rate-limiting considerations, and exposes these capabilities as discrete tools that agents can invoke.

Features

  • Global geocoding (forward and reverse)
  • Points-of-interest (POI) search across millions of places
  • Multi-modal directions (driving, walking, cycling) with traffic-aware routing
  • Travel time matrices for accessibility and logistics analysis
  • Route optimization (visit-order solving for multi-stop trips)
  • Map matching (snap GPS traces to roads)
  • Isochrone generation (areas reachable within time or distance limits)
  • Static map image generation (visualize locations, routes, markers)
  • Offline spatial utilities (distance, bearing, area, buffers, centroid)
  • Hosted MCP endpoint for quick access and local server option for private deployments
  • Integration guides for popular clients (Claude Desktop, VS Code, Cursor AI, etc.)

Installation / Configuration

Prerequisites:

  • Node.js (LTS)
  • A Mapbox access token (visit https://account.mapbox.com/)

Use the hosted endpoint (no install)

  • Endpoint: https://mcp.mapbox.com/mcp
  • Provide your Mapbox access token to your MCP client when configuring the endpoint.

Run locally (two common options)

  1. Run with npx (quick start)
# start server with npx, provide a Mapbox token via env var
MAPBOX_ACCESS_TOKEN=pk.your_token_here npx @mapbox/mcp-server
# default listens on http://localhost:3000/mcp
  1. Clone and run from source
git clone https://github.com/mapbox/mcp-server.git
cd mcp-server
npm install
export MAPBOX_ACCESS_TOKEN=pk.your_token_here
npm start
# or set PORT=8080 to change the listening port

Configuration notes

  • MAPBOX_ACCESS_TOKEN: required for any Mapbox-backed tools.
  • PORT: set to change the listening port (defaults to 3000).
  • The server exposes an /mcp endpoint that MCP clients can call. Consult your chosen client’s MCP configuration docs to point it at your local or hosted endpoint.

Available Tools / Resources

The MCP server packages Mapbox services into a set of tools agents can call. Below is a compact reference.

ToolPurpose
GeocodingForward/reverse address lookup and place search
POI SearchFind businesses, landmarks, and amenities (filter by category, proximity)
DirectionsTurn-by-turn routing with travel modes and traffic options
MatrixTravel time/distance matrices between sets of origins & destinations
OptimizationSolve visiting-order and routing optimization for multiple stops
Map MatchingSnap raw GPS traces to the road network
IsochronesGenerate polygons representing reachable areas by time/distance
Static MapsRender map images with markers, paths, and styles
Offline GeometryLocal calculations: distance, area, bearing, buffering, simplification

Primary repo and docs:

  • GitHub: https://github.com/mapbox/mcp-server
  • Hosted MCP Guide: see docs/hosted-mcp-guide.md in the repo
  • Integration guides: docs for Claude Desktop, VS Code, Cursor AI, Goose, and Smolagents

Use Cases

Concrete examples of what you can build or ask when the MCP server is available to your agent:

  • Local discovery
    • Prompt: “Find coffee shops within a 10-minute walk of the Empire State Building and return name, address, rating.”
    • Outcome: The agent invokes POI Search + Geocoding to return a structured list (JSON
Tags:locationsearch