OpenMetadata

OpenMetadata catalogs everything your data touches: tables, dashboards, pipelines, ML models, all searchable and lineage-tracked in one place. The project now calls itself an open context layer for data and AI, and that's not just marketing: recent releases added MCP support so AI assistants and agents can query the same catalog your team does. The core is Apache 2.0 and fully self-hostable, with no gated features.

Running it means Docker Compose or Kubernetes with real dependencies: Elasticsearch, MySQL or Postgres, and Airflow for ingestion. The 1.13 release was a big one: MCP as a first-class service category with SSO, a knowledge graph layer, a data marketplace, plus connectors for Google Drive, Pub/Sub, and the SAP estate. It also broke things: the Iceberg connector is gone, the Databricks connection scheme changed, and Python ingestion now needs 3.10+. Read the upgrade notes before touching a working install; a 2.0 release is in candidate builds now.

Solo devs won't need this. Small data teams finally learn what tables exist, who owns them, and whether they're fresh. Enterprise teams get governance and classification without buying Collibra or Atlan. Collate, the company behind it, sells a managed cloud with a free tier if you'd rather not run it.

The catch: it's a platform, not a tool. You're adopting an entire metadata layer, and a fast-moving one; major releases have earned their breaking-changes sections. If your data infra is two Postgres databases and a dbt project, this is overkill.