# 🌍 bbox-mcp-server
**Ask AI about anything, anywhere — and verify the answer on a map. For Free**
The geospatial toolkit for AI agents. **6 tools, zero config** — give any LLM the ability to find, query, convert, and aggregate spatial data using open data. No API keys or signups required to start.
Every response includes a shareable verification link to **[vibhorsingh.com/boundingbox](https://vibhorsingh.com/boundingbox/)** — click it to visually confirm results on an interactive map. No other MCP server does this.
[](https://nodejs.org/) [](./LICENSE)
---
## What Can You Do With This?
**You don't have to be a GIS professional to make use of this.** Any AI agent with this MCP server can answer spatial questions using real OpenStreetMap data and not spit out hallucinated garbage.
| Ask your AI agent... | What happens under the hood |
|---|---|
| *"How many EV chargers are in downtown Denver?"* | Overpass query → H3 hex binning → density analysis |
| *"Is there a hospital near this Airbnb?"* | POI search with structured tags → map verification link |
| *"Find all playgrounds within 1km of this address*"* | Overpass query + radius filter → pinned results on a shareable map |
| *"Convert this WKT to GeoJSON in EPSG:3857"* | Format conversion across 6 inputs, 9 outputs, 3,900+ projections |
| *"Show me all bike-share stations in Amsterdam"* | Curated OSM tags → Overpass query → results on map |
| *"Compare park density across Seattle neighborhoods"* | Overpass + H3 aggregation → hex-binned spatial analysis |
Every answer comes with a link. Click it, see if the AI got it right and you can also share it with someone else.
---
## What Makes This Different
Most AI tools give you text you have to trust. This one gives you an interactive map you can check and verify!
- **Verifiable** — Every response includes a public URL where you (or anyone) can visually confirm the results. This is the only MCP server that does this in any domain.
- **Deterministic** — Queries hit real OpenStreetMap data and return real coordinates. The AI interprets your question. The data comes from all the hard work done by all the awesome OSM volunteers.
- **Free** — No paid services. Overpass API is free, OSM data is free, the tool is free. A Mapbox token (free tier, no payment info) unlocks natural language location search but isn't required.
- **Zero config** — `npx -y bbox-mcp-server` and you're running.
---
## Why This Exists
| The problem | How bbox-mcp solves it |
|---|---|
| *"I have WKT but the API needs a GeoJSON bbox in EPSG:3857."* | Parses 6 input formats, projects to 3,900+ EPSG codes, outputs in 9 formats — in one call. |
| *"I keep getting the wrong OSM tags for Overpass queries."* | `list_osm_tags` returns curated tag combos. No more hallucinated `amenity=grocery`. |
| *"How many hospitals are in this district?"* | `aggregate_overpass_h3` queries Overpass and bins results into H3 hexagons server-side. |
| *"Is this bounding box actually correct?"* | Every response includes a clickable map link for visual verification. |
---
## Tools at a Glance
| Tool | What it does | Key params |
|---|---|---|
| `get_bounds` | Convert and project a bbox across formats and coordinate systems | `bbox`, `epsg`, `format`, `coord_order`, `zoom` |
| `get_h3_indices` | Generate H3 hex cell indices covering a bbox | `bbox`, `resolution`, `compact` |
| `generate_share_url` | Create a shareable map link for a bbox | `bbox` |
| `search_overpass` | Query OpenStreetMap via Overpass QL within a bbox or radius | `bbox`, `query`, `limit`, `radius_meters` |
| `list_osm_tags` | Look up correct OSM tags for a category | `category` |
| `aggregate_overpass_h3` | Run an Overpass query and bin results into H3 hexagons | `bbox`, `query`, `resolution` |
All tools accept `location` (natural language, requires Mapbox token) or `bbox` (coordinates, WKT, GeoJSON, etc).
---
## Quick Start
Add to your MCP client config:
Claude Desktop (Click to expand)
Add to `claude_desktop_config.json`:
```json
{
"mcpServers": {
"bbox": {
"command": "npx",
"args": ["-y", "bbox-mcp-server"]
}
}
}
```
Cursor (Click to expand)
Add to `.cursor/mcp.json`:
```json
{
"mcpServers": {
"bbox": {
"command": "npx",
"args": ["-y", "bbox-mcp-server"]
}
}
}
```
Windsurf (Click to expand)
Add to `~/.codeium/windsurf/mcp_config.json`:
```json
{
"mcpServers": {
"bbox": {
"command": "npx",
"args": ["-y", "bbox-mcp-server"]
}
}
}
```
VS Code (GitHub Copilot) or Google Antigravity (Click to expand)
Add to `.vscode/mcp.json` in your workspace:
```json
{
"servers": {
"bbox": {
"command": "npx",
"args": ["-y", "bbox-mcp-server"]
}
}
}
```
> Try: *"Find all coffee shops within 500m of Times Square"*
### Optional Configuration
```json
{
"mcpServers": {
"bbox": {
"command": "npx",
"args": ["-y", "bbox-mcp-server"],
"env": {
"MAPBOX_ACCESS_TOKEN": "pk.your-token-here",
"OVERPASS_API_URL": "https://your.custom.overpass.instance/api/interpreter",
"NOMINATIM_API_URL": "https://your.custom.nominatim.instance"
}
}
}
}
```
| Variable | Default | Description |
|---|---|---|
| `MAPBOX_ACCESS_TOKEN` | — | Enables natural language location search (e.g. *"San Francisco"*) |
| `OVERPASS_API_URL` | auto | Custom Overpass endpoint. By default, rotates between `overpass-api.de` and `kumi.systems`. |
| `NOMINATIM_API_URL` | `https://nominatim.openstreetmap.org` | Custom Nominatim endpoint. Set this if hosting as a shared/remote MCP server or using a commercial provider. |
| `MAX_H3_CELLS` | `50000` | Safety cap for H3 grid generation |
*Or install globally: `npm install -g bbox-mcp-server`*
---
## Tool Reference
### `get_bounds`
Convert and project a bounding box across 6 input formats, 9 output formats, and 3,900+ coordinate systems. Returns the center point and tile coordinates for the centroid.
| Param | Type | Default | Description |
|---|---|---|---|
| `bbox` | string | — | Input geometry (coordinates, WKT, GeoJSON, ogrinfo extent) |
| `epsg` | string | `"4326"` | Target projection. Unknown codes auto-fetched from epsg.io. |
| `format` | string | `"csv"` | Output: `csv`, `wkt`, `geojson-bbox`, `geojson-polygon`, `leaflet`, `overpass`, `ogc-bbox`, `kml`, `stac-bbox` |
| `coord_order` | string | `"lng,lat"` | Swap to `"lat,lng"` for APIs that expect latitude first |
| `zoom` | number | `15` | Zoom level for tile coordinate calculation |
| `precision` | number | `6` | Decimal places in formatted output |
**💡 Prompt:** *"Get the bounding box for Central Park in WKT format projected to EPSG:32618"*
---
### `get_h3_indices`
Generate Uber H3 hexagonal cell indices covering a bounding box.
| Param | Type | Default | Description |
|---|---|---|---|
| `bbox` | string | — | Input geometry |
| `resolution` | number | — | H3 resolution (0–15) |
| `compact` | boolean | `false` | Merge cells into coarser parents where possible |
| `return_geometry` | boolean | `false` | Include GeoJSON hex boundaries |
**💡 Prompt:** *"Give me H3 cells at resolution 7 for downtown Chicago, include the hex geometries"*
---
### `search_overpass`
Execute an Overpass query within a bounding box or radius. Returns structured results with names, coordinates, and tags. The verification link plots each result as a pin on the map.
| Param | Type | Default | Description |
|---|---|---|---|
| `bbox` | string | — | Input geometry |
| `query` | string | — | Overpass QL (e.g. `nwr["amenity"="cafe"]`). The server wraps it in the appropriate spatial filter automatically. |
| `radius_meters` | number | — | Search radius in metres. Use only when the user specifies an explicit distance (e.g. `2000` for "within 2km"). Omit for area/region queries. |
| `limit` | number | `100` | Max elements returned |
**💡 Prompt:** *"Find all parking within 2km of JFK airport"*
**💡 Prompt:** *"Search for hospitals in Manhattan, limit 50"*
---
### `list_osm_tags`
Look up the correct OpenStreetMap tags for a category before writing an Overpass query.
| Param | Type | Description |
|---|---|---|
| `category` | string | Broad category (e.g. `"food"`, `"health"`, `"transport"`) |
**💡 Prompt:** *"What are the correct OSM tags for supermarkets?"*
---
### `aggregate_overpass_h3`
Run an Overpass query and bin results into H3 hexagons for spatial density analysis. Returns counts per cell and GeoJSON hex boundaries.
| Param | Type | Default | Description |
|---|---|---|---|
| `bbox` | string | — | Input geometry |
| `query` | string | — | Overpass QL core query |
| `resolution` | number | `8` | H3 resolution for binning |
**💡 Prompt:** *"Aggregate all hospitals in Seattle into H3 bins at resolution 7"*
---
### `generate_share_url`
Generate a shareable link to visualize a bounding box on the interactive map at vibhorsingh.com/boundingbox.
| Param | Type | Description |
|---|---|---|
| `bbox` | string | Input geometry |
**💡 Prompt:** *"Generate a share link for bbox 40.7128,-74.0060,40.7580,-73.9855"*
---
## Supported Input Formats
All tools auto-detect the input format. No need to specify which one you're using.
| Format | Example |
|---|---|
| Raw coordinates | `40.7128,-74.0060,40.7580,-73.9855` |
| WKT | `POLYGON((-74.006 40.712, -73.985 40.712, ...))` |
| GeoJSON | `{"type":"Feature","geometry":{...}}` |
| GeoJSON bbox | `{"bbox":[-74.006,40.712,-73.985,40.758]}` |
| ogrinfo extent | `Extent: (-74.006, 40.712) - (-73.985, 40.758)` |
| Space-separated | `40.7128 -74.0060 40.7580 -73.9855` |
---
## 🤖 For AI Agent Developers
### Response structure
Every tool returns two content blocks:
1. **Human-readable text** — formatted output with the map verification link
2. **Structured JSON** — all computed data, machine-parseable
Example `get_bounds` JSON response:
```json
{
"original_wgs84": { "lat1": 40.7128, "lng1": -74.006, "lat2": 40.758, "lng2": -73.9855 },
"projected": { "xmin": -8238310.23, "ymin": 4970241.32, "xmax": -8235527.11, "ymax": 4976491.56 },
"center": { "lat": 40.7354, "lng": -73.99575 },
"tile_indices": { "z": 15, "x": 9660, "y": 12284 },
"epsg": "3857",
"coord_order": "lng,lat",
"area_km2": 8.681,
"dimensions": { "width_km": 1.714, "height_km": 5.066 },
"share_url": "https://vibhorsingh.com/boundingbox/#40.712800,-74.006000,40.758000,-73.985500"
}
```
### Error handling
All errors return `isError: true` with a descriptive message. Invalid coordinates, unknown EPSG codes, and oversized H3 requests all return clean errors — the server never crashes on bad input.
### Logging
Structured JSON logs go to **stderr** (stdout is reserved for MCP protocol). Each entry includes timestamp, level, and context.
---
## Acknowledgments & Fair Use
### OpenStreetMap
This tool is built on [OpenStreetMap](https://www.openstreetmap.org/): A global dataset created and maintained by millions of volunteers. Every query you run returns data that someone walked, mapped, or verified by hand. If you find this useful, consider [contributing to OSM](https://wiki.openstreetmap.org/wiki/How_to_contribute).
### Responsible Usage
The public Overpass and Nominatim API instances are free community resources with limited capacity. Avoid tight loops, excessive polling, or bulk-scraping. If hosting this as a shared/remote MCP server, point `NOMINATIM_API_URL` and `OVERPASS_API_URL` at your own instance or a commercial provider (see [Overpass API installation](https://wiki.openstreetmap.org/wiki/Overpass_API/Installation) and [Nominatim installation](https://nominatim.org/release-docs/latest/admin/Installation/)). The default server rotation for Overpass helps spread load, but it's not a substitute for responsible usage.
---
## License
MIT
[](https://lobehub.com/mcp/iamvibhorsingh-bbox-mcp-server)