BumiAtlas MCP

Tanzania: Zanzibar buildings by registration number

Zanzibar's building register as tools an AI assistant can call: find a building footprint by registration number or coordinates, read its area, and see it on a map with its neighbours.

Server URL

Streamable HTTP with OAuth. The first time you connect, your client opens a browser window to sign in with Google; no API key needed.

https://bumiatlas.com/mcp/tanzania

Connect

Claude (claude.ai, desktop)

Settings → Connectors → Add custom connector, paste the URL, then Connect and sign in with Google.

Claude Code

claude mcp add --transport http bumiatlas-tanzania https://bumiatlas.com/mcp/tanzania

ChatGPT

Settings → Apps & Connectors → Advanced → Developer mode, then Create with the URL and OAuth authentication.

Cursor

in ~/.cursor/mcp.json

{
  "mcpServers": {
    "bumiatlas-tanzania": {
      "url": "https://bumiatlas.com/mcp/tanzania"
    }
  }
}

Tools

Tool names and descriptions are in English, which models understand best; your assistant still answers in your language.

TZ_building_search

Find Zanzibar building footprints (COLA register) by building registration number or location. query takes a BRN ('120G40004') or coordinates as 'latitude, longitude' (e.g. -6.1659, 39.2026). With coordinates it returns buildings containing the point (distance 0), then the nearest within `radius` metres, closest first. Each building has id, brn, area (m², as registered), mapArea and bbox. Set include_geometry for each footprint as GeoJSON (keep limit small). Use TZ_building_image to see a building and its neighbours.

query*string: Building registration number (120G40004) or 'latitude, longitude'
radiusnumber: Search radius in metres around coordinates(default 25)
limitinteger: Maximum results(default 20)
include_geometryboolean: Include each boundary as a GeoJSON MultiPolygon(default false)

TZ_building_image

A map image centred on a Zanzibar building (id, from TZ_building_search) or a point (query), with every building footprint in view outlined and labelled with its BRN, so you can see which building is where. The building passed as id is drawn in red with a larger label; a query point is a red dot. Returns the buildings in view as JSON (label, id, brn, area in m²; labelled: false where its label didn't fit), then one JPEG per basemap. Basemaps: satellite (Esri World Imagery, often a few years old), hybrid (the same imagery with road and place names), streets and topo (Esri's street and topographic maps). Pass several to compare, e.g. ['satellite', 'streets'] to match buildings with street names. Footprint outlines are yellow on imagery and magenta on maps. Imagery is not always aligned with the register: offsets of a few metres are possible. A scale bar is at bottom left. Labels are left off where they'd overlap; call again with a smaller radius to read them. Each image costs about as much context as 1,000 words, so ask for what you need.

idstring: Building id (from TZ_building_search), drawn in red
querystring: 'latitude, longitude' to centre on instead of a parcel
radiusnumber: Metres from the centre to the image edge. Default: fits the parcel with a margin, or 80 m around a point
basemapsarray: One image per basemap(default ["satellite"])
sizeinteger: Image width and height in pixels(default 768)

Try asking

Which building is at -5.17008, 39.79258?
Look up building 92H20040.
Show the buildings around this point on satellite imagery.

About the data

~492k building footprints on Unguja and Pemba. Buildings, not land parcels; mainland Tanzania isn't covered.
A building registration number (BRN) can repeat on several structures of one registration.
No owners or occupants.