DataHub

What it is

Open-source metadata platform, now branded “The Context Platform for your Data and AI Stack” (GitHub description): discovery, governance, observability and context for humans and AI agents.

Maker, history

Originated at LinkedIn (2019, per the README); maintained by the DataHub project contributors and the DataHub company, which README says is founded by the creators of the open-source project. The managed MCP endpoint domain (*.acryl.io) and the acryldata GitHub organisation point to the company’s earlier name, Acryl Data; no rename date or funding verified, so no company note was created.

Editions and deployment

• DataHub Core: Apache 2.0, about 12.8k GitHub stars on 2026-10-07, pushed the same day.
• DataHub Cloud: managed, SLA-backed, full Context Platform plus observability. Pricing: vendor site.
• README claims 3,000+ organisations (Netflix, Visa, Etsy, Slack, Apple named; vendor claim).

Core architecture

Metadata graph ingested from 150+ integrations (README), column-level lineage, ownership, tags, glossary, domains, data products; access via GraphQL, API, SDK and MCP.

Role in an enterprise AI rollout

• Quality: assertions, freshness checks, SLA and incident tracking (Observability, README).
• Lineage: table and column level, upstream/downstream, also surfaced to agents.
• Access policy: metadata-level policies and ownership; does not enforce row/column data access itself (it is a catalog; opinion).
• PII handling: tags and glossary terms on columns, written via UI, ingestion or the MCP mutation tools.
• Semantics: Context Intelligence mines query history into a semantic index; Context Hub for expert review; Context Activation serves validated context to agents (README wording).

AI features as of October 2026

• Official MCP server (repo acryldata/mcp-server-datahub): search in plain English, inspect usage/ownership/docs/quality, trace table and column lineage, retrieve real SQL, review pending metadata proposals. Mutation tools (tags, glossary terms, owners, descriptions, proposals) need v0.5.0+ of the server. Managed endpoint on DataHub Cloud v0.3.12+; self-hosted via uvx. Auth: OAuth2 with dynamic client registration (Cloud v1.0.2+) or personal access tokens / service accounts.
• Context Platform capabilities above are vendor-described; per-feature GA/preview status not stated.

Integrations

Warehouses, BI, orchestration and transformation tools via ingestion; used alongside dbt, Airflow, Iceberg catalogs. Compared in data-catalogs-compared.

Strengths and weaknesses (opinion)

Strong open-source community and ingestion breadth; good agent access via MCP. Self-hosting needs operational effort; governance workflows in the open edition are lighter than commercial suites such as collibra.

Self-learning

• Docs: https://docs.datahub.com/ (free); MCP guide: https://docs.datahub.com/docs/features/feature-guides/mcp
• Repo: https://github.com/datahub-project/datahub
• Certifications: none verified.

Sources

https://github.com/datahub-project/datahub, https://raw.githubusercontent.com/datahub-project/datahub/master/README.md, https://docs.datahub.com/docs/features/feature-guides/mcp, GitHub API repo metadata (all fetched 2026-10-07).

Open items

• Rename from Acryl Data (date), funding, current release version, GA status of Context Platform features.