Mapbox MCP is Mapbox's official server for exposing geospatial APIs through the Model Context Protocol. The mapbox/mcp-server repository describes a Node.js server with geocoding, place search, routing, travel-time matrices, isochrones, map matching, route optimization, static images, interactive map rendering, and offline geospatial calculations. In the GitHub check on August 26, 2026, the repository was MIT licensed, had 352 stars and 41 forks, and the @mapbox/mcp-server npm package was at version 0.14.0.
For Educasium, Mapbox MCP has a precise role: it is a geospatial intelligence connector for assistants that need to reason about a site, an address, an access area, or a map. It does not replace QGIS for heavier GIS work or EnergyPlus for energy simulation. It is mainly useful for turning a spatial question into coordinates, routes, reachable zones, and visual evidence.
Contents
- What Mapbox MCP does
- Installation, hosted endpoint, and requirements
- Search, routing, and accessibility tools
- Interactive maps, static images, and resources
- Comparison with QGIS and EnergyPlus
- Security, cost, and limits
- Educasium position
What Mapbox MCP Does
Key point: Mapbox MCP gives an AI client tools to understand where places are, how to connect them, and how to show the result on a map. The server combines Mapbox API calls, which depend on token rights and quotas, with offline geospatial calculations based on Turf.js.
Capability families
The verified README presents Mapbox MCP as an MCP layer over Mapbox APIs. The main capabilities are global geocoding, point-of-interest search, driving, walking, and cycling routes with real-time traffic, travel-time matrices, route optimization, map matching for GPS traces, isochrones, and map rendering.
The value for a school, architecture office, real estate team, or logistics team is the combination. An assistant can start from an ambiguous address, ask the user to disambiguate if several places match, find nearby facilities, compute a 15-minute walking area, then render a map. Doing that manually across several interfaces would be slower and easier to mis-document.
What is out of scope
Mapbox MCP is not a full GIS. It does not replace complex layer management, advanced topology, attribute joins, professional map layout, or QGIS processing workflows. It also does not replace a regulatory study, traffic engineering study, or thermal simulation.
It should be described as an access and visualization connector. For urban, site-selection, or logistics decisions, it can prepare assumptions and exploratory maps, but conclusions still need professional review and source verification.
Installation, Hosted Endpoint, And Requirements
Key point: the repository offers a hosted endpoint at https://mcp.mapbox.com/mcp, and the npm package currently requires Node 22 or newer. A Mapbox access token is required for Mapbox API-backed tools.
Two usage modes
The README first offers a hosted MCP endpoint. That is the fastest trial path because the MCP client connects directly to the Mapbox URL. The user still needs a Mapbox access token from a Mapbox account. This is not a small detail: without a token, the API-backed tools cannot work correctly.
The repository also supports local execution through @mapbox/mcp-server. The checked package.json lists version 0.14.0, MIT license, a mcp-server binary, and a Node engine requirement of 22 or higher. The dependency on @modelcontextprotocol/ext-apps explains the richer in-chat map previews for clients that support MCP Apps.
Clients and integrations
The repository includes guides for Claude Desktop, Goose, VS Code, Cursor, and Smolagents. It also documents importing the tools directly into an application, which matters when a team wants Mapbox capabilities inside its own agent rather than inside a desktop client.
The experience is not identical in every client. MCP resources and MCP Apps are not universally supported. The repository therefore includes resource fallback tools for clients that cannot read MCP resources natively.
Search, Routing, And Accessibility Tools
Key point: Mapbox MCP is strongest when the question involves access, proximity, travel time, and visual comparison. These are common in real estate, architecture, education, mobility, and logistics.
Search and geocoding
Search and geocoding turn text into usable places. The server groups place search, geocoding, reverse geocoding, category search, and place details. In an Educasium scenario, a learner could ask for facilities around a campus, services near a parcel, or shops reachable from an address.
The README also documents ground_location_tool for coordinate-first questions. It chooses between neighborhood context, nearby POI search, reachable region, and closest routable point depending on the query. That keeps a simple question from turning into a fragile chain of manual tool calls.
Routing, matrices, and isochrones
Directions cover several travel profiles, including driving, traffic-aware driving, walking, and cycling. Matrices compute travel times and distances between many points. Isochrones show what can be reached from a point within a specified time or distance.
For site decisions, this is more useful than straight-line distance. Two parcels five kilometers from a center may have very different accessibility because of the road network, barriers, traffic, or travel mode. Mapbox MCP helps the assistant make those differences visible.
Interactive Maps, Static Images, And Resources
Key point: render_map_tool is the main visual primitive, while static_map_image_tool is the reliable fallback for clients without interactive rendering. The distinction matters because it changes what the user can inspect.
Interactive rendering
The README explains that render_map_tool uses the MCP Apps protocol to display a Mapbox GL JS map inside supported clients. The clients named include Claude Desktop, VS Code with GitHub Copilot, Claude Code, and Goose. The same tool can display results produced by Mapbox tools or manually composed GeoJSON.
This is valuable in teaching. An interactive map lets the user zoom, inspect, compare, and correct. In a demonstration, the assistant can generate an initial map, then the learner can ask for a different travel profile, radius, satellite style, extra markers, or a custom polygon.
Images and resources
static_map_image_tool produces a base64 PNG or JPEG. It is less flexible than an interactive map, but more portable. For a page, report, or MCP client without MCP Apps, the static image is the stable path.
The server also exposes resources such as mapbox://categories and mapbox://categories/{language}. These give the available categories for place search. For clients without MCP resource support, CLIENT_NEEDS_RESOURCE_FALLBACK=true exposes equivalent tool calls.
Comparison With QGIS And EnergyPlus
Key point: Mapbox MCP answers where, QGIS MCP handles detailed GIS work, and EnergyPlus MCP handles simulated building performance. They are complementary connectors.
| Need | Mapbox MCP | QGIS MCP | EnergyPlus MCP |
|---|---|---|---|
| Find an address or POI | Strong fit | Possible with imported data | Out of scope |
| Compute routes and isochrones | Strong fit | Possible with plugins/data | Out of scope |
| Manage complex GIS layers | Limited | Strong fit | Out of scope |
| Simulate a building | Out of scope | Spatial context only | Strong fit |
| Render a fast map in chat | Strong fit | Possible via QGIS rendering | Out of scope |
Picking the right connector
For a quick site-selection study, Mapbox MCP is often the first step. It is fast for proximity, accessibility, and visualization. If the study requires cadastral layers, zoning, geoprocessing, or controlled map layouts, QGIS MCP becomes the better tool.
If the question is about energy, comfort, or building performance, Mapbox is not enough. It can contextualize the location, but EnergyPlus MCP takes over for loading an IDF, inspecting zones and materials, changing loads, running simulations, and analyzing outputs.
Security, Cost, And Limits
Key point: Mapbox MCP handles coordinates, addresses, geometries, and an API token, so confidentiality, quotas, and professional context matter. Sensitive spatial data does not become harmless because an assistant processed it.
Data sent to APIs
The README's privacy section says Mapbox API requests can include addresses, search text, coordinates, route preferences, waypoints, vehicle dimensions, map parameters, markers, and geographic data. For local execution, the README states that calls go directly from the user's environment to Mapbox APIs and that the local MCP server does not store the requests. The hosted endpoint at https://mcp.mapbox.com/mcp adds a Mapbox-operated intermediary and should be evaluated under the hosted service's terms.
That does not remove professional caution. A confidential site address, logistics route, or intervention area may still be sensitive. Professional users should check Mapbox terms, privacy policy, and internal rules before sending that data.
Quotas and correctness
Online tools depend on Mapbox APIs, token rights, and account quotas. Offline calculations do not necessarily consume API calls, but they depend on the quality of the geometry provided. Wrong coordinates or units can still produce a convincing but false map.
Educasium should make assumptions visible: address used, selected coordinates, travel mode, departure time, units, and source. Without those details, the map looks more certain than it is.
Educasium Position
Key point: Educasium should present Mapbox MCP as assisted spatial reasoning, not as a final authority. The pedagogical value is strong when assumptions and limits are explicit.
What to teach
A good Mapbox MCP workshop starts with a concrete case: compare two sites for a campus, check walking access around an address, map nearby services, or visualize a 20-minute driving area. The goal is not to impress with a map. The goal is to show how AI turns a vague request into verifiable spatial data.
The exercise should include a control step. The learner verifies the chosen place, compares the result with an official map or GIS source, reads the routing assumptions, and documents uncertainty. That discipline keeps automation separate from validation.
Next step
The positioning should stay sober. Mapbox MCP is official, Mapbox-maintained, and functionally rich, but it relies on external APIs and account conditions. Reliable use cases, technical requirements, token handling, and complementarity with QGIS and EnergyPlus matter more than a promise of automatic mapping.