OpenMetadata

What it is

Open-source metadata and catalog platform, described on GitHub as “The Open Context Layer for Data and AI”: trusted data context and business semantics for humans, AI assistants and agents. It unifies technical metadata, data quality signals, lineage, ownership and governance in one knowledge graph.

Maker, history

Maintained by Collate, which sells the commercial platform (managed enterprise capabilities, AI agents, automation, AI Studio, enterprise MCP workflows, support). Founding dates, funding and the relationship history are not verified; no Collate company note was created.

Editions and deployment

Open source, Apache 2.0, about 15.4k GitHub stars on 2026-10-07, pushed the same day. Self-hosted, or managed via Collate (pricing: vendor site).

Core architecture

Central metadata store with a unified schema, 130+ connectors (README), table, column, dashboard, pipeline and metric lineage; data quality test cases, profiling, freshness checks and observability signals built in; data contracts.

Role in an enterprise AI rollout

• Quality: native test cases and profiler, so quality lives next to the catalog.
• Lineage: multi-level lineage readable by agents through MCP.
• Access policy: roles/policies for metadata; not a data-plane enforcer (opinion).
• PII handling: classification and tags, auto-classification features exist (not detailed in sources fetched).
• Semantics: glossary, data contracts, semantic search; “memory” so agents can preserve organisational knowledge.

AI features as of October 2026

• MCP server ships installed and enabled by default (Settings > Applications); search metadata, inspect lineage, data contracts, a semantic (vector) search tool. Auth: OAuth 2.0 (recommended), personal access token, bot token for unattended agents (docs v2.0.x paths).
• AI SDK for governed access to context; Collate-only: AI agents, AI Studio, enterprise MCP workflows (README wording).
• GA/preview status of each not stated.

Integrations

Warehouses, BI, pipelines and messaging via connectors; can sit beside dbt and orchestrators. Compared in data-catalogs-compared; closest peer is datahub.

Strengths and weaknesses (opinion)

All-in-one open source (catalog plus quality plus lineage) with simple operations; MCP on by default is convenient but needs scoped bots. Smaller commercial ecosystem than the large suites.

Self-learning

• Docs: https://docs.open-metadata.org/ (free); MCP: https://docs.open-metadata.org/latest/how-to-guides/mcp
• Repo: https://github.com/open-metadata/OpenMetadata
• Certifications: none verified.

Sources

GitHub README and API metadata, MCP docs page (fetched 2026-10-07).

Open items

• Collate history/funding, current release number, auto-classification details.