M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
M

Mapbox MCP Server

Official
Bring Mapbox's mapping, geocoding, and routing intelligence into your AI assistant
How to install ↓ Compare GitHub source repository ↗
Category
Other #8 of 230
Stars
★ 361 Very popular
Transport
stdio (local process) · Streamable HTTP
Runtime
Node.js · Hosted (no local runtime)
Credentials
API key / credential required
License
MIT
Last commit
Tools
31
71FMRS · B

Officially maintained by Mapbox, this MCP server bundles geocoding, POI search, routing, isochrones, map rendering, and a broad set of offline geometry tools into one comprehensive service, but it depends heavily on a Mapbox access token and network connectivity, and sends location data to Mapbox APIs.

Strongest · Documentation 17/20 Weakest · Reliability 11/20

Reliability
11/20
Security and permissions
14/20
Maintenance
16/20
Documentation
17/20
Setup experience
13/20
Why each score
Reliability 11/20
The manifest declares both an npx/stdio local package and a remote streamable-http endpoint, and the README's tool descriptions (geocoding, routing, isochrones, offline geometry, etc.) are detailed and consistent with the manifest description with no obvious contradictions. However, this review is based only on the README and manifest as static evidence — no CI workflow files or committed test files were included to verify behavior; the `npm test`/`npm run lint` commands mentioned in CONTRIBUTING are process claims only, not verifiable execution evidence. Per the static calibration rule, reliability is capped at 12 absent such evidence, so 11 is given: the happy path is plausible from the description, but edge cases, error handling, and dependency control remain unverified in the supplied material.
Security and permissions 14/20
Credential handling is reasonably well disclosed: MAPBOX_ACCESS_TOKEN is marked required+secret, and the README's 'Data Usage & Privacy' section states the token stays local and is never stored/logged by the server, while listing exactly what data each tool category sends to Mapbox APIs — a decent level of data-boundary transparency. All tools are read-only geospatial queries/calculations (geocoding, routing, isochrones, offline turf.js math, etc.); no destructive, payment, or remote-execution defaults are present, so no red line is triggered. That said, gaps remain: the manifest also declares an official hosted remote endpoint (`https://mcp.mapbox.com/mcp`) whose authentication mechanism (whether it reuses the same token, and how it's transmitted) is not described in the material, leaving the trust boundary for that path unclear; additionally, `render_map_tool` renders custom HTML/GeoJSON panels via the MCP Apps protocol, and its sandboxing/permission isolation is not detailed. This does not amount to a red-line violation, but permission/confirmation disclosure is incomplete, warranting 14.
Maintenance 16/20
The repository is officially maintained by Mapbox, carries a clear MIT license, and the manifest version (0.14.0) is paired with a documented release process in the README (version bump, manifest sync, changelog prep, CI validation of version consistency), showing solid release governance. The README explicitly states a 'Maintenance Commitment' promising ongoing updates, bug fixes, and MCP protocol compatibility, and lists two support channels (email and GitHub Issues). The repo is not archived, and has 350 stars with only 4 open issues, which is only a secondary signal of responsiveness (not used to add points directly). Since no actual commit timeline or recent release dates were available to verify sustained activity, it falls short of a top score — 16 is given.
Documentation 17/20
Documentation is layered and thorough: the main README details each tool's function, parameters, and usage examples, and links out to multiple client integration guides (Claude Desktop, VS Code, Cursor, Goose, Smolagents) as well as dedicated docs for resources, elicitations, tracing, and the render_map_tool payload schema, plus an explicit data usage/privacy section. Example prompts span discovery, navigation, visualization, and offline geometry use cases, and a deprecated tool (category_list_tool) is clearly flagged with a migration path. The gap is that concrete API cost/rate-limit details are not included in the supplied material (only links to external Mapbox docs), so it falls short of a perfect score — 17 is given.
Setup experience 13/20
Two setup paths are offered: using the officially hosted endpoint (no local install) or running locally via `npx @mapbox/mcp-server` with the `MAPBOX_ACCESS_TOKEN` environment variable, with clear linked steps for obtaining a token. The README also links to several client-specific configuration guides. However, per the static calibration rule, setup is capped at 15 absent verifiable CI/test evidence supporting a reproducible working connection; also, the supplied README excerpt does not itself include a copy-pasteable JSON client config snippet (configuration is spread across external linked docs), adding some friction. 13 is given.

Static review · not runListed 2026-08-26

Read the FMRS scoring method →

Fit and risk

What it can accessRuns commands or codeUses the network

Best for

  • Developers building AI applications that need real geocoding, POI search, and routing data
  • Logistics, travel, and site-selection workflows that need offline geometry operations (area, buffers, union/intersect/difference)
  • Users on clients with MCP Apps support (Claude Desktop, VS Code Copilot, Claude Code, Goose) who want visual, interactive map output

Not for

  • Users without a Mapbox account who don't want to obtain an access token
  • Fully offline geocoding or routing needs, since most location APIs require live Mapbox connectivity (only the Turf.js geometry tools work offline)
  • Fleet-scale optimization with time windows, capacity constraints, or multiple vehicles — the V2 Optimization API needs beta access and isn't registered by default

Required permissions

  • Requires a valid Mapbox access token (MAPBOX_ACCESS_TOKEN), supplied as a secret environment variable
  • Local stdio deployment needs outbound network access to call Mapbox APIs
  • Using the hosted endpoint requires network access to https://mcp.mapbox.com/mcp

Risks and side effects

  • A leaked Mapbox access token can be abused, incurring extra API charges or exceeding quota
  • Geocoding, search, and directions tools send addresses, coordinates, and routing preferences to Mapbox APIs, subject to Mapbox's privacy policy
  • render_map_tool renders an HTML panel via the MCP Apps protocol, which relies on the client's handling of rich content being trustworthy
  • The advanced V2 Optimization API ships in the codebase but is unregistered by default, requiring deliberate beta opt-in

Setup

Before you start

Runtime：Node.js · Hosted (no local runtime)

MAPBOX_ACCESS_TOKEN requiredsecret Required Mapbox access token. Sign up for a free account at mapbox.com, then create a new token or use the default public token on the Account page at account.mapbox.com.
Other optional settings (1)
CLIENT_NEEDS_RESOURCE_FALLBACK optional Optional. Set to true for clients without MCP resource support (like smolagents) to enable the resource_reader_tool and category_list_tool fallback tools; leave unset by default.
  1. Sign up for a Mapbox account at mapbox.com/signup and create (or use the default) access token from your account page.
  2. For local use, add @mapbox/mcp-server to your MCP client config to be launched via npx, setting the MAPBOX_ACCESS_TOKEN environment variable.
  3. Alternatively, point your client directly at the hosted endpoint https://mcp.mapbox.com/mcp (streamable-http) without running anything locally.
  4. If your client lacks native MCP resource support (e.g. smolagents), set CLIENT_NEEDS_RESOURCE_FALLBACK=true to enable fallback tools.
  5. Restart the client and test with a geocoding or directions query to confirm connectivity.
claude_desktop_config.json 
{"mcpServers": {"mapbox": {"command": "npx", "args": ["-y", "@mapbox/mcp-server"], "env": {"MAPBOX_ACCESS_TOKEN": "YOUR_MAPBOX_ACCESS_TOKEN"}}}}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "mapbox": {
      "command": "npx",
      "args": [
        "-y",
        "@mapbox/mcp-server"
      ],
      "env": {
        "MAPBOX_ACCESS_TOKEN": "YOUR_MAPBOX_ACCESS_TOKEN"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add mapbox -e MAPBOX_ACCESS_TOKEN=YOUR_MAPBOX_ACCESS_TOKEN -- npx -y @mapbox/mcp-server

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After configuring the client, ask "What's the distance between San Francisco (37.7749, -122.4194) and New York (40.7128, -74.0060)?" — a numeric distance answer proves the connection works, and Mapbox API tools like directions_tool should appear in the client's tool list (offline geospatial tools work without the token).

Troubleshooting

  1. Verify MAPBOX_ACCESS_TOKEN is set correctly, unexpired, and scoped for the APIs being used
  2. If tool calls fail, check network access to api.mapbox.com or mcp.mapbox.com
  3. If your client doesn't support MCP resources, set CLIENT_NEEDS_RESOURCE_FALLBACK=true to enable resource_reader_tool and category_list_tool
  4. If map rendering doesn't appear, confirm the client supports MCP Apps (Claude Desktop, VS Code Copilot, Claude Code, Goose); otherwise use static_map_image_tool for a static image
  5. For version-related issues, confirm package.json, manifest.json, and server.json versions are in sync

Things to try

Once connected, you can ask your AI assistant things like:

  • Find coffee shops within walking distance of the Empire State Building
  • Get driving directions from LAX to Hollywood with current traffic
  • Show me areas reachable within 30 minutes of downtown Portland by car
  • Optimize a delivery route for these 8 addresses: 1600 Pennsylvania Ave, 350 5th Ave, 1 Infinite Loop, 2300 Truxtun Rd, 4150 Tempest St, 600 B St, 275 5th Ave, 501 W Broadway

Tools 31

resource_reader_tool read-only
Fallback tool that lets clients without native MCP resource support read resources such as the category list.
category_list_tool read-only
Deprecated tool that returns the category ID list; use resource_reader_tool or the native resource API instead.
distance_tool read-only
Computes the great-circle distance between two coordinates offline using the Haversine formula, in multiple units.
points_within_polygon_tool read-only
Tests one or more points against a polygon or multipolygon, returning only those inside — useful for geofencing and delivery-zone checks.
destination_tool read-only
Calculates a destination point from a starting coordinate, bearing, and distance using a geodesic offset, without any API call.
bearing_tool read-only
Calculates the compass bearing (0-360°) and cardinal direction from one coordinate to another.
midpoint_tool read-only
Finds the geographic midpoint between two coordinates along the great-circle path.
centroid_tool read-only
Calculates the centroid of a polygon or multipolygon.
Show 23 more tools
area_tool read-only
Calculates the area of a polygon in units such as square meters, square kilometers, or acres.
bounding_box_tool read-only
Calculates the minimum bounding box containing a point, line, polygon, or multipolygon.
buffer_tool read-only
Creates a buffer polygon around a point, line, or polygon at a given distance.
simplify_tool read-only
Reduces the vertex count of a line or polygon using the Douglas-Peucker algorithm.
length_tool read-only
Measures the total length of a line defined by a series of coordinates.
convex_tool read-only
Computes the convex hull — the smallest convex polygon containing a set of points.
nearest_point_tool read-only
Finds the nearest point in a collection to a given target point.
nearest_point_on_line_tool read-only
Snaps a point to the nearest position on a line or route and returns the distance.
union_tool read-only
Merges two or more polygons into a single geometry.
intersect_tool read-only
Finds the overlapping area shared by two or more polygons.
difference_tool read-only
Subtracts one polygon from another to compute the remaining area.
matrix_tool read-only
Calls the Mapbox Matrix API to compute travel-time and distance matrices between multiple points, with traffic-aware departure times.
static_map_image_tool read-only
Calls the Mapbox Static Images API to generate static map images with markers, routes, and polygon overlays.
category_search_tool read-only
Calls the Mapbox Search Box category search API to find points of interest by category, filtered by proximity.
reverse_geocode_tool read-only
Calls the Mapbox Geocoding V6 API to convert coordinates into human-readable addresses and location context.
ground_location_tool read-only
Answers 'what's near this coordinate' in one call by combining reverse geocoding, category search, and isochrone, returning cited results.
place_details_tool read-only
Retrieves detailed information for a place by Mapbox ID, including photos, hours, ratings, phone number, and website.
directions_tool read-only
Calls the Mapbox Directions API for driving, walking, and cycling routes, with waypoints, traffic, and exclusion options.
isochrone_tool read-only
Calls the Mapbox Isochrone API to compute areas reachable within a given travel time or distance.
search_and_geocode_tool read-only
Calls the Mapbox Search Box Text Search API to search for and geocode POIs, addresses, and places.
map_matching_tool read-only
Calls the Mapbox Map Matching API to snap noisy GPS traces onto the road network.
optimization_tool read-only
Calls the Mapbox Optimization API to solve the optimal visiting order across multiple stops (TSP).
render_map_tool read-only
Renders a live, interactive Mapbox GL JS map directly inside supported chat clients via the MCP Apps protocol, showing routes, search results, or custom GeoJSON.

Use cases

Building location-aware AI travel assistants or place recommendation apps
Computing travel-time matrices and optimizing multi-stop delivery routes for logistics
Cleaning up and visualizing GPS traces snapped to the road network
Analyzing reachable areas within a given time or distance (isochrones) for site selection or coverage analysis
Rendering interactive or static maps of routes and locations directly inside a chat session

Supported clients

Claude Desktop
Claude Code
GitHub Copilot
Goose
CursorPartial support
SmolagentsPartial support

Listed from the project's documentation, not tested by this site.

Overview

Mapbox MCP Server is a Node.js server officially maintained by Mapbox that exposes geocoding, POI search, multi-modal directions, travel-time matrices, route optimization (TSP), GPS map matching, isochrone analysis, static and interactive map rendering, and offline Turf.js-based geospatial calculations (distance, area, bearing, buffers, union/intersect/difference, etc.) as MCP tools and resources. It can run locally over stdio via npm, or connect directly to Mapbox's hosted streamable-http endpoint; either path requires a Mapbox access token.

Similar servers

BioMCP 75 · B

One binary. One grammar. Evidence from the biomedical sources you already trust.

★ 653 · Tools 11 Compare with this →
Compare with other MCP servers

Source revision 77fecbd21394 Data synced 2026-10-11 Read the FMRS scoring method