
it’s easy to try out. I had a local install running quickly on Docker and could test a lot with it.
Governance features: Glossaries with approval workflows, classifications and tags, tiers, certification (Bronze/Silver/Gold), domains and data products are all built in. This covers what a Data Steward needs day to day.
Clear permission model: Access is built from Policies, Roles, Teams and Users. When an action is blocked, the error names the missing operation (for example EditAll or ResolveTask). That made it easy to work out which role changes we need.
Collaboration inside the tool: Owners, @mentions, tasks, conversations and notifications let stewards raise and follow up on issues next to the asset itself.
Lineage and data quality: Lineage, test cases and incidents sit in the same place as the metadata.
MCP integration: OpenMetadata exposes its metadata through MCP, so an AI assistant can work with the catalog directly. In our testing it could:
search assets and read entity details (owners, tags, domains, descriptions, glossary terms);
read lineage;
create and update entities (descriptions, domains, glossaries);
check what a change did, independently of the UI.
We used this to verify UI test results, for example to confirm which tables actually had a glossary term. That is much faster than opening each asset by hand. It also lets the catalog be queried in natural language, so people can use it without knowing the UI.
Practical extras: CSV import/export, powerful search, and custom properties. Review collected by and hosted on G2.com.
Steward role is too restricted out of the box: The built-in Data Steward policy only allows editing descriptions, tags, owners and similar fields. A steward could not create glossaries or terms, build term hierarchies, resolve or reassign tasks, create classifications and tags, or create test cases and data products. We had to write a separate requirements sheet with 32 changes, 15 of them role/policy fixes.
Some permissions are unclear: Several actions failed with no error message, so we had to guess which permission was missing. Some errors are also hard to read. For example, EditAll is needed even for simple glossary edits like adding related terms.
Missing features:
no abbreviations field on glossary terms;
no Experts option on tables;
no Follow option on glossary terms;
no description field on user profiles;
no “mark all as read” for notifications;
no filter for assets with or without a glossary term, or with no domain.
Weak guardrails:
a glossary term with an empty definition is accepted without warning;
any certification level can be applied to any asset, with no check on quality;
the history doesn’t show what changed when a certification is downgraded.
Admin area visible to stewards: A steward can see the admin area and every user, which may not be wanted.
Notifications: The owner is not notified of a plain conversation on an asset. It only works with an @mention. SMTP is disabled by default, so email alerts need extra setup. Alerts also have no Test Case source in this version.
Search index lag: New or changed descriptions and tags sometimes took a while to show in search, and tag results mix table-level and column-level tags.
Setup and stability: On our Docker setup, the ingestion scheduler and the source database container stopped, which blocked some tests until we restarted them. Some form errors are unhelpful, such as “query param category must not be null” when saving an alert, or “Invalid date format: null” for a Date custom property.
Backend bug: One table had an owner pointing to a deleted glossary term reference, which caused errors. Review collected by and hosted on G2.com.