🌍 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 — click it to visually confirm results on an interactive map. No other MCP server does this.

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.

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

Tools at a Glance

All tools accept location (natural language, requires Mapbox token) or bbox (coordinates, WKT, GeoJSON, etc).

Quick Start

Add to your MCP client config:

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "bbox": {
      "command": "npx",
      "args": ["-y", "bbox-mcp-server"]
    }
  }
}

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "bbox": {
      "command": "npx",
      "args": ["-y", "bbox-mcp-server"]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "bbox": {
      "command": "npx",
      "args": ["-y", "bbox-mcp-server"]
    }
  }
}

Add to .vscode/mcp.json in your workspace:

{
  "servers": {
    "bbox": {
      "command": "npx",
      "args": ["-y", "bbox-mcp-server"]
    }
  }
}

Try: "Find all coffee shops within 500m of Times Square"

Optional Configuration

{
  "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"
      }
    }
  }
}

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.

💡 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.

💡 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.

💡 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.

💡 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.

💡 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.

💡 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.

🤖 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:

{
  "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: 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.

Responsible Usage

The public Overpass API instances are free community resources with limited capacity. Avoid tight loops, excessive polling, or bulk-scraping. For heavy workloads, set OVERPASS_API_URL to your own instance, see Overpass API installation. The default server rotation helps spread load, but it's not a substitute for responsible usage.

License

MIT