The Amazon DataZone data catalog is AWS’s business catalog for a data estate: publish an asset into a domain, attach glossary terms and metadata forms, and route access through a subscription approval that lands the real permission in AWS Lake Formation or Amazon Redshift. One governance model now runs on two product versions, domainVersion V1 and V2, across two domain identity modes. Which combination you run decides what the catalog can do.
A V2 domain opens through Amazon SageMaker Unified Studio, where AWS calls the same catalog Amazon SageMaker Catalog, “built on Amazon DataZone.” (retrieved 2026-09-16), and both names stay current. Above either version sits an enterprise context layer, the category Atlan builds, holding definitions that outlive a console. The split between this catalog and the metastore under it is worked out in AWS DataZone vs Glue Data Catalog, the baseline in what a data catalog is, and the distinction in data catalog vs context layer.
| Quick facts | Amazon DataZone data catalog, as of 2026-09-16 |
|---|---|
| What it is | AWS’s business data catalog: inventory, curate, publish, subscribe, inside a domain |
| Official names | Amazon DataZone, and Amazon SageMaker Catalog in SageMaker Unified Studio |
| The two versions | domainVersion V1 and V2. Upgrade is console-only, reversible only before any SageMaker-created project |
| The two identity modes | IDC-based and IAM-based. 2026 catalog features shipped per mode |
| Access enforcement | AWS Lake Formation and Amazon Redshift, managed assets only |
| Agent interface | HTTPS API. No AWS-supported MCP server for this catalog |
Which Amazon DataZone do you have, V1 or V2, IDC or IAM?
Two axes decide your capability set, both belonging to the domain rather than the service. Version is first: domainVersion V1 opens the Amazon DataZone data portal, V2 opens Amazon SageMaker Unified Studio. Three signals tell you which: the portal your users open, whether the AmazonSageMakerDomainExecution and AmazonSageMakerDomainService roles exist, and the domainVersion value itself, which cannot be misread. AWS’s upgrade documentation (retrieved 2026-09-16) notes CloudFormation templates need editing afterward.
Identity mode is second; through 2026, catalog capabilities arrived per mode, not all at once. The catalog user guide is itself titled “Catalog in IDC-based domains”. Data lineage in IAM-based domains shipped 2026-07-07 and custom asset types 2026-07-09, per the release notes; cross-Region subscriptions for IDC-based domains posted 2026-01-20. Two constraints interact: IAM roles cannot log in to SageMaker Unified Studio, an AWS Important note, and under AWS Identity Center single sign-on, a domain must sit in the same Region as that instance.
A V1 domain that is only IAM-owned needs a human or SSO owner before upgrading, or the result is unusable; a V1 domain that is SSO-owned should have its environment profiles inventoried first, since project profiles must re-express all of them. On V2, IDC-based domains get cross-Region subscriptions, posted 2026-01-20, while IAM-based domains need their Region’s 2026 releases confirmed before anyone designs against them.
One honest limit, checked 2026-09-16: AWS publishes no per-capability GA-versus-preview matrix for SageMaker Catalog, and no sentence in the user guide assigns either label. Release dates are all there is. What the catalog governs once the version is settled sits in Amazon DataZone vs Atlan, with the cross-cloud read in data lineage for AI.
The changelog gap: November 2024 on one page, August 2026 on the other
AWS keeps two dated release histories that touch this catalog, and their newest entries are worth reading. The “What is new in Amazon DataZone?” page covers 2024 and 2023 only, newest entry dated 20 November 2024. The SageMaker Unified Studio release notes run to 31 August 2026, with the catalog appearing month after month. A third dated fact belongs beside those: AWS’s pricing page states “As of Nov 1, 2024, there is no monthly user-based subscription charge for Amazon DataZone.” Set the three dates side by side and read them.
Now the other side: a changelog is evidence about where a vendor publishes, not much more. Amazon DataZone became generally available on 4 October 2023. It runs 19 Regional endpoints, AWS added Hong Kong, Malaysia and Zurich on 23 September 2025, HIPAA eligibility has been in the changelog since 14 December 2023, and every DataZone API is unchanged in Unified Studio. The shared console still offers “Create an Amazon DataZone domain”, and nothing on docs.aws.amazon.com/datazone/ says otherwise, checked 2026-09-16.
The useful question is which version you are on and what the move costs, more answerable than whether the data catalog is finally dead. A business’s vocabulary for its own data moves on its own schedule, the distinction active metadata vs context layer draws.
What does an upgrade to SageMaker Unified Studio carry, and what does it drop?
The upgrade changes one domain property in place. Most of what you built crosses untouched; two entity classes do not. The mechanics, from AWS’s upgrade documentation and the June 2025 launch note, both retrieved 2026-09-16: console only, no API support, two new roles, and rollback to V1 only while no SageMaker-created project exists, with the AWS Resource Access Manager permission staying behind on the way back. One quieter effect: the Amazon Q subscription drops to the free tier, which matters if your users reached the catalog through the Amazon Q, QuickSight and Glue bundle or Amazon Quick Suite.
Then there is the one-way door. AWS states it plainly, retrieved 2026-09-16:
“Environments and environment profiles from Amazon DataZone will not show in Amazon SageMaker Unified Studio - these entities have been replaced by Amazon SageMaker project profiles. Projects created in the Amazon SageMaker Unified Studio will not be visible through the Amazon DataZone portal.”
Environments and environment profiles are how a project gets its Athena workgroup, Glue database and Redshift connection, so the compute wiring of every project has to be re-expressed as project profiles. The invisibility runs both ways, leaving a partially migrated estate with two catalogs and neither showing the whole picture.
Where the Amazon DataZone data catalog runs out of room
Subscription approval creates the grant for you in AWS Lake Formation or Amazon Redshift, and AWS’s concepts documentation (retrieved 2026-09-16) scopes that automatic fulfillment to managed assets: AWS Glue tables and Amazon Redshift tables and views. For any other asset type, AWS publishes an Amazon EventBridge event with the details, you create the grant yourself wherever the data lives, then call updateSubscriptionStatus to close the request: the precise edge of what approval means here.
One AWS-internal contradiction deserves naming, not quiet resolution. The concepts page above still says that “in the current release of Amazon DataZone, you can create and run data sources for AWS Glue and Amazon Redshift,” listing no third-party source, while AWS’s Snowflake data source page documents a Snowflake source in procedural detail. Both are live, both AWS’s own, and this page reports the disagreement rather than settling it.
The documented limits land early. A run fails after 100 tables, JDBC, DocumentDB and MongoDB are unsupported sources, both from AWS’s lineage documentation, with column-level detail depending on source data. Data-source runs cap at 25 per source per day against adjustable ceilings of 1,000,000 assets and 10,000 glossary terms, per the quota table. What that 100-table ceiling costs an agent asking an impact question sits in automated SQL lineage vs manual lineage mapping and AWS Glue Data Catalog limitations.
| Layer | Authoritative for | Where its authority ends |
|---|---|---|
| AWS Glue Data Catalog | Tables, schemas, partitions | No business meaning, no approval workflow |
| DataZone / SageMaker Catalog | Glossary, forms, publishing, subscription requests | Does not hold the permission |
| Lake Formation and Redshift | The grant on managed assets | Unmanaged assets route through EventBridge to you |
| IAM | Account and Organization identity | Knows nothing about business definitions |
Reach across clouds is a separate, narrower axis: AWS’s pricing page describes DataZone as cataloging data “stored across AWS, on premises, and third-party sources.” That question is treated in AWS DataZone vs a cross-cloud data governance platform and AWS data governance.
How do agents reach the Amazon DataZone data catalog today?
An agent reaches this catalog by one of four routes, and as of 2026-09-16 none is an AWS-supported MCP server built for it. The DataZone HTTPS API is the durable route, largely undisrupted by the version change. AWS publishes no stage label on that surface, only the service’s general availability, announced 4 October 2023.
The AWS MCP Server is now generally available, in AWS’s own headline, posted 6 May 2026: managed, remote, IAM SigV4 auth, and a call_aws tool that executes any of the 15,000-plus AWS API operations under the caller’s credentials. It is generic, with no DataZone-specific or SageMaker-Catalog-specific tools; an agent reaches the catalog through it by calling the API, which is what when to use MCP vs an API turns on. Amazon Q inside Unified Studio gives documented natural-language search over SageMaker Catalog.
The Agent Toolkit route needs its qualifier read carefully. Its aws-data-analytics plugin bundles catalog skills, making AWS’s own catalog route a skills route rather than a server route, the distinction in agent skills vs MCP. Those skills read the AWS Glue Data Catalog’s business context, and AWS’s announcement calls that capability a preview in its own headline, posted 17 June 2026, in four Regions, with a SearchAssets API in AWS’s getting-started guide. Still preview as of 2026-09-16, and no sentence in either document says otherwise. The fourth route sits in the awslabs GitHub organization, whose own README settles its standing: “This is an unofficial, community-developed project and is not affiliated with, endorsed by, or supported by Amazon Web Services, Inc.”
An HTTPS API is not an MCP server, and the generic AWS MCP Server is not catalog tooling. Both statements are narrow, and both matter when an agent needs business context at runtime, the gap an MCP-connected data catalog closes and MCP delivering business context explains. Every status above carries its date because these move, the AWS Agent Registry included.
Coverage or answers: which one tells you the catalog is working?
A catalog’s health shows up two ways, and AWS meters only one. Coverage is the traditional measure: assets documented, terms defined, users onboarded. A catalog serving agents has to be measured by answers instead: did it find what it needed, and if not, why not.
Look at what AWS bills for. Its pricing page meters requests at $10 per 100,000 (4,000 free monthly), metadata storage at $0.4 per GB (20 MB free), compute at $1.776 per unit (0.2 free), and generative-AI recommendations at $0.015/$0.075 per 1,000 input/output tokens with no free tier. Search is a free API, exempt from the request meter along with the domain, environment, project, user and policy management groups. Searching costs nothing. Storing costs money.
| Meter | What AWS counts | Coverage or answer? |
|---|---|---|
| Requests | Billable API calls, Search and management groups excluded |
Coverage |
| Metadata storage | Size of domains, glossaries, terms, forms, projects, users, assets | Coverage |
| Compute | Metadata ingestion and recommendation processing | Coverage |
| Generative-AI recommendations | Input and output tokens for generated descriptions | Coverage |
| An agent’s answer | Not metered | Neither. Not a number AWS publishes |
Whether an agent got its answer appears on neither page, so the scoreboard a team inherits from its bill is a coverage scoreboard: the documented route to a program that reports green while its users route around it, the failure mode why data governance implementations fail walks.
Search alone does not close the gap. AWS describes the search algorithm as prioritizing keyword matches, then appending semantic ones: keyword first, semantics second. A vectorized context layer serves four modes instead: vector, semantic, hybrid and graph traversal, over MCP, SQL and APIs, the mode that answers the impact questions a keyword index cannot reach, the separation knowledge graph vs data catalog makes. On that arc, metadata supply here is Automated and consumption Collaborative, ahead of most enterprise catalogs. The missing third column is learning: usage traces routing back into better context, which nothing in the documented product does, the gap where a data catalog for AI and types of metadata AI agents need both start.
How does Atlan keep definitions stable across a DataZone version change?
A definition typed into a vendor console inherits that console’s lifecycle. Whichever version you run, the glossary term has to mean the same thing to an agent AWS did not build. Atlan is the enterprise context layer: the place a definition lives so a domain upgrade moves the console, not the meaning. Four components carry it: Context Agents create context from evidence rather than waiting on a curation backlog, the Context Lakehouse stores and vectorizes it in open tables in your own cloud, MCP and conversational AI serve it to agents and people through one connection, and a learning loop routes failed answers back into creation so a gap gets filled instead of logged.
AWS documents the pairing itself. Its SageMaker Unified Studio user guide states that “the integration between Amazon SageMaker Catalog and Atlan enables bidirectional metadata synchronization across both platforms,” per AWS Documentation, retrieved 2026-09-16: on-demand and scheduled sync, glossary terms with parent-child relationships, ingestion of projects, assets, domains, data products and forms, and real-time reverse sync into SageMaker Catalog. AWS’s own third-party catalog integrations page names third-party business data catalogs, Atlan among them, as the route for business context above SageMaker Catalog. Where that context gets created is covered in AI agents for a data catalog and a context catalog; build-or-buy criteria sit in build vs buy for metadata tooling and context layer vs data catalog vs semantic layer.
The Amazon DataZone data catalog is a real enforcement point: access requests route through it, and approval lands a grant in Lake Formation or Redshift. It is less suited to holding the definitions your agents ask questions against, because the console those definitions were typed into has already changed name and version once, and two of its entity types did not make the trip. A definition that has to outlive a product version has to live somewhere that is not that product.
Real customers, real stories: Modern data catalog in action
53 % less engineering workload and 20 % higher data-user satisfaction
"Kiwi.com has transformed its data governance by consolidating thousands of data assets into 58 discoverable data products using Atlan. 'Atlan reduced our central engineering workload by 53 % and improved data user satisfaction by 20 %,' Kiwi.com shared. Atlan's intuitive interface streamlines access to essential information like ownership, contracts, and data quality issues, driving efficient governance across teams."
Data Team
Kiwi.com
🎧 Listen to podcast: How Kiwi.com Unified Its Stack with Atlan
One trusted home for every KPI and dashboard
"Contentsquare relies on Atlan to power its data governance and support Business Intelligence efforts. Otavio Leite Bastos, Global Data Governance Lead, explained, 'Atlan is the home for every KPI and dashboard, making data simple and trustworthy.' With Atlan's integration with Monte Carlo, Contentsquare has improved data quality communication across stakeholders, ensuring effective governance across their entire data estate."
Otavio Leite Bastos, Global Data Governance Lead
Contentsquare
🎧 Listen to podcast: Contentsquare's Data Renaissance with Atlan
FAQs about the Amazon DataZone data catalog
1. Can I still create a new Amazon DataZone domain in 2026?
Yes. No deprecation, end-of-support or sunset notice appears anywhere on AWS’s DataZone documentation as of 2026-09-16. The service runs 19 Regional endpoints, three added 23 September 2025, and the shared SageMaker console still offers “Create an Amazon DataZone domain.”
2. What is the difference between a DataZone V1 domain and a V2 unified domain, and how do I tell which I have?
The domainVersion property on the domain: V1 opens the Amazon DataZone data portal, V2 opens Amazon SageMaker Unified Studio, where the same catalog appears as SageMaker Catalog. Three checks answer it: which portal your users open, the domainVersion value, and whether the two SageMaker domain roles exist.
3. What does an upgrade to SageMaker Unified Studio not carry over?
Environments and environment profiles, replaced by Amazon SageMaker project profiles. It runs in reverse too: projects created in SageMaker Unified Studio are not visible through the Amazon DataZone portal. Rollback is available only while no SageMaker-created project exists, and CloudFormation templates need edits.
4. What is the difference between an IDC-based and an IAM-based domain for the catalog?
Identity mode, which gates catalog features. Data lineage in IAM-based domains shipped 2026-07-07 and custom asset types 2026-07-09, while cross-Region subscriptions for IDC-based domains posted 2026-01-20. AWS marks one constraint Important: IAM roles cannot log in to SageMaker Unified Studio.
5. Is there an official MCP server for the Amazon DataZone data catalog?
No AWS-supported one, as of 2026-09-16. Agents reach it through the generic AWS MCP Server, generally available since 6 May 2026 with its call_aws tool, through Amazon Q inside SageMaker Unified Studio, or through Agent Toolkit skills reading the Glue catalog’s business context, still labeled a preview in four Regions. The awslabs community server disclaims AWS affiliation in its own README.
Sources
All sources retrieved 2026-09-16.
- DataZone concepts
- What is new in Amazon DataZone?
- Upgrade a domain
- Create DataZone domains
- SageMaker vs DataZone
- Data lineage
- Snowflake data source
- Search for data
- Endpoints and quotas
- DataZone API Reference
- Unified Studio release notes
- Catalog in IDC-based domains
- Third-party catalog integrations
- Atlan integration
- Create a DataZone domain from the SageMaker console
- Business context in the Glue Data Catalog
- DataZone pricing
- DataZone general availability
- Domain upgrade to SageMaker
- Three additional Regions
- Glue business context, preview
- Cross-Region subscriptions
- AWS MCP Server general availability
- Secondary source:
awslabs/amazon-datazone-mcp-serverREADME, for its own unofficial-project disclaimer.
