Setting Up the Amazon Quick Agentic Catalog
On this page
Quick answer
The Amazon Quick Agentic Catalog Experience is a preview curator workflow for AWS Glue Data Catalog and Databricks Unity Catalog. Connect one supported catalog, choose Explore data, scope the Quick Agent to that source, describe the business use case, review its recommendations, and create DirectQuery Datasets plus a multi-dataset Topic.
The flow has four documented stages:
- Discover: describe the use case and let the agent recommend relevant tables from catalog metadata.
- Create: review the recommendations, then create a focused set of DirectQuery Datasets in bulk.
- Relationships and Topics: review inherited or inferred relationships and create a Topic spanning the chosen assets.
- Semantic inheritance: bring table and column descriptions into the generated Datasets.
AI recommendations are not approvals. AWS explicitly tells authors to review discovered tables, inferred relationships, and inherited descriptions.
Do not confuse two Amazon Quick catalog paths
Amazon Quick now has a separate integration with the GA AWS Agent Registry. It is not the Agentic Catalog Experience described by this page.
| Quick path | What it discovers | What it enables |
|---|---|---|
| Agentic Catalog Experience | AWS Glue or Databricks catalog metadata | Curator review, DirectQuery Datasets, relationships, Topics, and inherited semantics |
| AWS Agent Registry integration | Agents and MCP servers from an organization’s registry | Search, browse, enable with pre-populated connection details, and share across chat, agents, apps, flows, and deep research |
The registry integration is configured from the Quick admin console under Manage account → Permissions → AWS Agent Registry. Its six-Region overlap footprint and agent/MCP permissions are separate from the data-catalog preview’s catalog, identity, DirectQuery, and semantic-governance decisions.
Choose the catalog and authentication path
| Catalog | Discovery authentication | Optional per-user enforcement |
|---|---|---|
| AWS Glue Data Catalog | Service role or AWS IAM Identity Center | Trusted identity propagation through IAM Identity Center and Lake Formation |
| Databricks Unity Catalog | Personal Access Token or OAuth 3LO | OAuth 3LO identity propagation |
Identity propagation is optional and is not required for discovery, bulk creation, or semantic inheritance. Read the identity propagation guide before choosing it.
For a detailed source decision, use the Glue versus Databricks comparison.
Connect and scope the source
Create the supported data source in Amazon Quick. For Glue, AWS’s walkthrough also creates and links an Athena data source: Glue supplies catalog metadata and Athena supplies the query path to S3 data.
Choose Explore data from the catalog data source. Confirm that the Agent opens in Specific data mode with the intended connection pinned as the context boundary. Do not expose the entire enterprise catalog when one business domain is sufficient.
Before discovery, verify:
- catalog, database, schema, and table access;
- source and query-engine connectivity;
- the service role, PAT, OAuth, or Identity Center path;
- which curator can create and share Quick assets;
- preview acceptance and a rollback plan.
Discover and review assets
Describe the business user, decision, metrics, time grain, and required dimensions. Ask the Agent to summarize the relevant part of the catalog and show relationships.
Review every recommended table against the upstream catalog. Check owners, descriptions, quality status, freshness, sensitivity, row grain, keys, and metric definitions. A relevant-sounding table name is not proof that it is approved for the use case.
Create only the intended subset. AWS recommends a focused context boundary because thousands of unrelated tables add noise to dashboards and AI Q&A.
Create Datasets and a Topic
After approval, create the selected Datasets in bulk. Catalog-Generated Datasets are DirectQuery representations by default and display a Semantics inherited indicator.
Then review the proposed Topic:
- confirm primary and foreign keys;
- inspect cardinality and join direction;
- test star or snowflake relationships with known totals;
- reject ambiguous many-to-many joins;
- verify metric aggregation and time dimensions;
- approve the Topic only after reconciliation.
See the generated Datasets and Topics guide for the asset boundary.
Validate before sharing
Run fixed questions whose correct answers are known. Compare totals and joins against the source, test users with different permissions, and inspect query logs. Include ambiguous language, missing fields, stale descriptions, and adversarial catalog text.
Do not describe the preview as production-ready merely because one demonstration works. Define an owner for connection failures, semantic changes, permission drift, bad joins, sync, and user feedback.
Frequently asked questions
Which catalogs support the Amazon Quick agentic catalog preview?
AWS currently documents AWS Glue Data Catalog and Databricks Unity Catalog for agentic discovery, bulk dataset creation, and semantic inheritance.
Does the Quick Agent publish datasets without review?
Authors control the workflow and should review discovered tables, inferred relationships, and inherited descriptions before proceeding.
Does the setup copy data into Quick?
Catalog-Generated Datasets use DirectQuery by default. They query data at its source; the upstream catalog remains authoritative for inherited metadata.
Is this the AWS Agent Registry integration for agents and MCP servers?
No. This guide covers the preview data-catalog curator flow. The separate Agent Registry integration lets Quick users discover, enable, and share registered agents and MCP servers.
Official sources
- Amazon Quick: The agentic catalog experience
- Amazon Quick: Supported catalogs
- AWS launch post: Agentic Catalog Experience
- AWS: Agent Registry agents and MCP servers in Amazon Quick
- AWS: Agent Registry generally available
Source check: September 2, 2026. The data-catalog flow remains a distinct preview; verify its catalogs and permissions separately from Agent Registry integration, Regions, and terms.