Catalogs and ownership: who owns this table and is it safe to use?

At a glance

Learning objectives

• Run the catalog assistant's four-step workflow on a real asset
• Answer owner, SLA and classification questions from metadata alone
• Keep catalog access read-only and metadata-only

What you need

• A DataHub catalog with owners, tags and classifications populated
• The data-catalog-assistant agent wired to mcp-datahub

Steps

1. Name the job — an agent that answers "what is this data, who owns it, and is it safe to use?". (from agent-data-catalog-assistant )
2. Follow the workflow — resolve the assets from the catalogue, retrieve schema, ownership and classification, retrieve lineage and freshness, then summarise with provenance and caveats. (from agent-data-catalog-assistant )
3. Know the outputs — an asset summary, owner and SLA, classification, and upstream/downstream. (from agent-data-catalog-assistant )
4. See it answer — 'which tables feed the revenue mart and who owns them?', with owners and freshness. (from agent-data-catalog-assistant )
5. Respect the guardrails — read-only, and never returns raw row data, only metadata. (from agent-data-catalog-assistant )
6. Ground it in the model — the open metadata model for ownership and lineage, ingestion framework, search, and governance integration. (from ref-datahub-docs )
7. Secure the path — read-only against the catalog, but catalogue metadata can itself be sensitive; use a scoped token and keep the GMS endpoint internal. (from mcp-datahub )

Check yourself

• Asked "which tables feed the revenue mart and who owns them?", the answer names tables, owners and freshness — not guesses.
• The assistant returns metadata only — if an answer contains row values, the guardrail was violated.

Worked example

Q: which tables feed the revenue mart and who owns them?
A: per asset — summary, owner & SLA, classification, upstream/downstream
Guardrails: read-only; never returns raw row data, only metadata.

Sourced from

• Data Catalog Assistant
• DataHub Documentation
• DataHub MCP Server

Security scan

Last sweep 2026-10-24: 0 finding(s), 0 marked as examples.

Common gotchas

• A catalog without owners is a faster search engine — populate ownership before you need it.
• Classification is part of the answer, not garnish — it decides who may see the data at all.
• The GMS token is scoped for a reason — a shared admin token turns a read-only assistant into a liability.

Production considerations

• Point the assistant at the same catalog your lineage tools read — one source of truth for ownership.
• Audit quarterly that every alerting asset has an owner and an SLA in the catalog.

Diagram

catalog assistant workflow
   |
   |- 1. resolve asset(s) from the catalogue
   |- 2. retrieve schema, ownership, classification
   |- 3. retrieve lineage and freshness
   |- 4. summarise with provenance and caveats
   |
    --> outputs: asset summary | owner & SLA | classification | upstream/downstream

Registry stats