Marmot

Marmot provides a searchable catalog of data assets—including tables, topics, queues, and APIs—augmented with lineage, ownership, business definitions, and custom fields. It aggregates metadata from a range of sources such as PostgreSQL, MySQL, MongoDB, ClickHouse, BigQuery, Kafka, Airflow, dbt, S3, DuckDB, and SQS, presenting it through a unified UI, API, and a built‑in MCP server that AI agents can query for up‑to‑date context.

The platform is designed for both engineers and AI‑driven tools, allowing users to discover assets in seconds, trace data dependencies, and retrieve domain‑specific explanations (e.g., term definitions). Metadata can be populated via Terraform, Pulumi, API, or CLI, and the MCP endpoint enables any MCP‑compatible client to ask natural‑language questions backed by the catalog.

Marmot runs as a single binary backed only by PostgreSQL, minimizing infrastructure requirements. It can be started with a simple Docker Compose command, scales to hundreds of thousands of assets and dozens of concurrent users while keeping average response times under 50 ms, and is positioned as an experimental, open‑source alternative to traditional data catalog platforms.