Quick Start

The hosted MCP server is available at:

https://duckdb-mcp.nrp-nautilus.io/mcp

Add it to your LLM client and start asking questions about the data.

VS Code

Create a .vscode/mcp.json in your project:

{
  "servers": {
    "duckdb-geo": {
      "url": "https://duckdb-mcp.nrp-nautilus.io/mcp"
    }
  }
}

Claude Desktop

Add to your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "duckdb-geo": {
      "url": "https://duckdb-mcp.nrp-nautilus.io/mcp"
    }
  }
}

Restart Claude Desktop after saving.

Example questions

Once connected, ask your LLM client questions like:

• What fraction of Australia is protected area?
• How many km² of wetlands overlap with vulnerable carbon stores?
• Which watersheds have the highest biodiversity scores?

The agent will call browse_stac_catalog and get_stac_details to discover dataset paths, then execute DuckDB SQL against S3.

MCP tools

Tool	Description
browse_stac_catalog	List all available datasets from the STAC catalog
get_stac_details	Get S3 paths and column schemas for a dataset
query	Execute DuckDB SQL against S3 Parquet files

MCP resources & prompts

Some clients (Claude Code, Continue.dev) also support MCP resources and prompts:

• catalog://list — list all datasets
• catalog://{dataset_id} — schema for a specific dataset
• geospatial-analyst prompt — load the full analyst persona and context

Local development

Run the server locally against your own CPU and network:

pip install -r requirements.txt
python server.py

Then connect via http://localhost:8000/mcp (note: http, not https).

WARNING

Local queries use your machine's CPU and network. Large S3 scans will be significantly slower than the hosted k8s endpoint.