What it does OpenMetadata is an open-source metadata platform that ingests technical and operational metadata from 120+ data services—warehouses, dashboards, pipelines, ML platforms—and connects it all into a unified knowledge graph. It tracks column-level lineage, data quality test results, ownership, glossary terms, classifications, and policies in one place.

The interesting bit The project is explicitly built for the AI assistant era. It ships an MCP server so LLMs and agents can query the metadata graph through natural language, inspect lineage, and understand business semantics (“this cust_id means Customer, not Account”) rather than blindly connecting to raw databases. The README frames this as “AI does not need another raw database connector. AI needs context.”

Key highlights

• 120+ connectors for databases, lakes, dashboards, pipelines, messaging, ML platforms, and more
• Column-level lineage for precise impact analysis across tables, dashboards, and ML models
• Governed business semantics: glossaries, metrics, domains, data products, and classifications (PII, GDPR, HIPAA, etc.)
• Built-in MCP server for AI assistant/agent integration via semantic search and natural language queries
• Open metadata standards with JSON Schemas, RDF/OWL ontologies, SHACL shapes, and JSON-LD contexts

Caveats

• The README is heavily AI-marketed; the actual MCP server capabilities are only partially listed before truncation (“create glossaries an…”)
• No specific performance, scale, or deployment complexity numbers are provided in the sources

Verdict Worth evaluating if you’re running a multi-tool data stack and want LLMs or data teams to stop guessing what buyer_key actually means. Probably overkill if your “data estate” is three Postgres tables and a Grafana dashboard.