dbt DuckDB MCP Claude Free

Build a data pipeline.
Give it an AI interface.

A hands-on lab. You write real dbt models on National Park Service data, run 18 automated tests, then connect Claude so anyone can query it in plain English.

Start the Lab → View on GitHub

What you'll build

1

Load and clean real data

Seed 5,824 species and 23,296 observations from four national parks into DuckDB. Write staging models that deduplicate and validate the source data.

2

Build analysis-ready tables

Write mart models that answer real questions — which park has the most wildlife activity, how many endangered species exist, which animals are most observed.

3

Test and document everything

Run 18 not_null and unique tests with dbt test. Generate a schema catalog that documents every column.

4

Connect Claude to the database

Wire up a read-only MCP server so Claude can list tables, read schemas, and run SELECT queries. Ask questions in plain English, get real answers back.

Data: NPS Biodiversity Dataset — Bryce, Great Smoky Mountains, Yellowstone, and Yosemite. Real numbers: Yellowstone has 1,443,562 recorded observations.