Kartoza Spatial MCP
kartoza/kartoza-mcp
Análise geoespacial nativa para seu LLM
Ver no GitHubServidor MCP de análise geoespacial de alto desempenho que integra clustering DBSCAN com aceleração Quadtree para o Claude e outros clientes MCP. Processa pontos GeoJSON e retorna clusters como polígonos prontos para mapa, em um único binário estático.
Instalação
npx convez-ai kartoza/kartoza-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Clustering de pontos geográficos em fluxos de IA
- Análise espacial sem depender de scripts Python externos
- Integração de operações GIS em agentes Claude e Gemini
Veredito

Solução bem-construída para quem precisa de análise geoespacial real em workflows de LLM. Binário estático, MCP-compliant e focado — ótimo para DevOps e desenvolvedores que querem evitar Python no pipeline.
README
A tradução para o português deste README ainda está na fila.
TL;DR — One static Go binary. Speaks MCP over stdio. Runs DBSCAN with a Quadtree on GeoJSON points and returns the clusters as map-ready polygon circles. Drop it into Claude Desktop or Gemini CLI and you're done.
Why you might want this
| If you're a… | …this is for you because |
|---|---|
| User | You want your LLM to actually do spatial analytics, not write Python that maybe runs somewhere. |
| Developer | You want a tiny, well-shaped place to add buffers, hulls, voronoi, isochrones — without re-doing MCP plumbing. |
| DevOps | You want a single static binary that you can package, sandbox, observe and forget. |
📖 Full documentation: kartoza.github.io/kartoza-mcp
Quickstart
git clone https://github.com/kartoza/kartoza-mcp.git
cd kartoza-mcp
nix develop # reproducible toolchain
nix run .#build # build the static binary
nix run .#test # green suite on first run
./spatial-cluster-static < test.json # smoke test
Then wire it into Claude Desktop:
{
"mcpServers": {
"kartoza-mcp": {
"command": "nix",
"args": ["run", "github:kartoza/kartoza-mcp#run"]
}
}
}
Restart Claude Desktop. Ask: "What MCP tools do you have available?" —
you should see cluster_points.
Features
- DBSCAN clustering with Quadtree acceleration —
O(N log N)neighbour search. - Parallel neighbour pre-pass via goroutines.
- Static binary built with
musl— drop on any Linux box, no shared libraries. - MCP-compliant with protocol version
2024-11-05. - REUSE / SPDX-compliant. Every file headered. CI keeps it honest.
- Brand-aligned mkdocs site under the Kartoza palette and motif.
Project shape
kartoza-mcp/
├── main.go # MCP server + cluster_points handler
├── main_test.go
├── flake.nix # Reproducible dev shell + apps
├── docs/ # mkdocs site (published to gh-pages)
├── mkdocs.yml
├── SPECIFICATION.md # Authoritative product spec
├── PACKAGES.md # Annotated dependency list
├── AGENT.md # Brief for AI assistants
└── README.md
Roadmap
The first tool is cluster_points. Next on the bench:
convex_hull— smallest enclosing polygon.buffer— fixed-distance buffers around features.voronoi— Voronoi diagram from a point cloud.nearest_neighbour— k-NN with optional weight functions.
Open an issue if you'd like one of those (or something different) prioritised.
Documentation
The mkdocs site at kartoza.github.io/kartoza-mcp is organised around three audiences:
- 🧭 Getting Started — install, first cluster, Claude Desktop config.
- 👤 For Users — tool reference, example prompts, visualisation, troubleshooting.
- 💻 For Developers — architecture, adding a tool, coding standards, testing.
- 🛠 For DevOps — deployment, containers, systemd, observability, security.
- 📚 Reference — MCP protocol, schema, edge cases.
Sponsor this work
Open source is sustained by people who choose to fund it. If
kartoza-mcp saves you time or sells you a contract, consider
sponsoring on GitHub or
hiring Kartoza for bespoke work.
License
MIT — REUSE-compliant. Every source file carries SPDX headers. See docs/about/license for the third-party bill of materials.