--- title: "AtScale and Unity Catalog: What It Reads, What It Doesn't" url: "https://atlan.com/know/ai-agent/databricks/atscale-unity-catalog/" description: "See exactly what AtScale's Unity Catalog integration governs on Databricks, what it doesn't, and where a context layer closes the gap for AI agents." author: "Emily Winks" author_role: "Data Governance Expert" published: "2026-09-07" updated: "2026-09-07T00:00:00.000Z" --- AtScale is a third-party semantic layer, not a Databricks-built one, and it reads Unity Catalog the way any well-behaved outside tool should: through tables, schemas, and grants, without moving data or reinventing access control. Atlan sits a layer above that connection, governing the business glossary and cross-platform lineage neither AtScale nor Unity Catalog owns. For the general question of what Unity Catalog's own semantic layer is, see [Unity Catalog's semantic layer](https://atlan.com/know/ai-agent/semantic-layer/unity-catalog-semantic-layer/). This page stays narrower: what changes when AtScale sits on Unity Catalog, where that governance runs out, and where Atlan picks up. AtScale recently shipped an MCP server, listed on the Databricks MCP Marketplace since December 1, 2025, that lets AI agents call its metrics under Unity Catalog's access control. Crediting that plainly, then naming exactly where it stops, is the point of what follows. - AtScale imports Unity Catalog tables and views as fact and dimension datasets over a Databricks SQL connection, no data movement required. - Unity Catalog enforces row and column-level security on AtScale's behalf, rather than AtScale rebuilding that layer itself. - AtScale's MCP server inherits Unity Catalog's access control and auditability for agent calls against its metrics. - By its own description, AtScale is "not a data catalog nor a policy engine": a glossary, cross-platform lineage, and unstructured knowledge sit outside both products. --- ## Integration overview | Attribute | Details | |---|---| | Integration | AtScale (third-party universal semantic layer) reading Databricks Unity Catalog | | Mechanism | AtScale imports Unity Catalog tables and views as fact and dimension datasets over a Databricks SQL warehouse connection, with no data movement | | What AtScale reads from Unity Catalog | Catalogs, schemas, tables, and object-level grants | | What Unity Catalog enforces on AtScale's behalf | Row-level security and column masking, delegated rather than rebuilt | | AI agent access (MCP) | AtScale's MCP server, listed on the Databricks MCP Marketplace since December 1, 2025; agent calls to AtScale's metrics inherit Unity Catalog's access control, auditability, and traceability | | What Unity Catalog doesn't supply | A business glossary, cross-platform lineage, unstructured knowledge, or a policy layer, by AtScale's own words, "not a data catalog nor a policy engine" | | Where Atlan fits | Governs and compounds context across AtScale's Databricks slice and every other warehouse or tool, alongside AtScale rather than in place of it | --- ## Is AtScale the same thing as Unity Catalog's own semantic layer? Two different products answer to a similar-sounding name, and a reader searching "AtScale Unity Catalog" could be looking for either. AtScale is a third-party semantic layer vendor, one of several profiled in [Atlan's roundup of semantic layer tools](https://atlan.com/know/best-semantic-layer-tools/), that predates and runs independently of Unity Catalog's own metrics features; a team can run AtScale, Databricks' native Unity Catalog Metrics, both, or neither. For the broader question of what Unity Catalog's own semantic layer is against dbt and Cube, see [Unity Catalog's semantic layer](https://atlan.com/know/ai-agent/semantic-layer/unity-catalog-semantic-layer/). For the Databricks-native metrics story feeding Genie Ontology, see [Unity Catalog Metrics](https://atlan.com/know/ai-agent/databricks/unity-catalog-metrics/). The genuine complexity worth naming: Unity Catalog's own Pages feature (Beta) lets a team author a governed business definition that [Genie Ontology](https://atlan.com/know/ai-agent/databricks/genie-ontology/) prioritizes over inferred context, while AtScale's semantic model is a separate, unconnected source of business meaning in the same Databricks stack. Neither vendor treats the other's definitions as authoritative. A metric can mean one thing in an AtScale cube and something adjacent in a Unity Catalog Pages entry, and nothing reconciles the two, the predictable result of two governed slices sitting side by side without a shared layer above them. AtScale gets called a catalog, a semantic layer, and a context layer in the same sentence often enough that [semantic layer vs. data catalog](https://atlan.com/know/ai-agent/semantic-layer/semantic-layer-vs-data-catalog/) exists to sort the three apart. For the direct comparison, AtScale against a governed context layer rather than the platform question here, see [AtScale vs a governed metadata and context layer](https://atlan.com/know/ai-agent/semantic-layer/atscale-vs-context-layer/). --- ## What does AtScale actually read from Unity Catalog? AtScale's Databricks integration connects over a Databricks SQL warehouse connection and imports Unity Catalog tables and views as fact and dimension datasets, the raw materials its semantic models are built from. According to [Databricks' own Unity Catalog documentation](https://docs.databricks.com/aws/en/data-governance/unity-catalog/), Unity Catalog operates as "the unified governance layer for data and AI built into Databricks," covering access control, lineage, and auditing automatically across workspaces, which is the governance surface AtScale's connection rides on rather than replaces. AtScale reads Unity Catalog's object-level grants and leans on its native row-level security and column masking to enforce access, instead of maintaining a parallel permissions system. That's real, useful delegated governance: a query against an AtScale metric respects the same row and column rules a query against the underlying table would. AtScale's Semantic Modeling Language, the open-source spec behind its models, sits at 173 stars on [GitHub](https://github.com/semanticdatalayer/SML), Apache-2.0 licensed, engineer-oriented adoption rather than a groundswell, but a sign the modeling layer is meant to be inspectable, not a black box. What this connection doesn't carry across is anything outside Databricks. AtScale's read of Unity Catalog is real and worth crediting, technical metadata and delegated enforcement inside one platform, but it stops at that platform's edge. Atlan's [Enterprise Data Graph](https://atlan.com/know/enterprise-data-graph/) is the counterpart to what doesn't travel with it: column-level lineage reverse-engineered from SQL across every connected system, not just the one warehouse AtScale happens to be reading from that day. The same boundary holds one level up. [Atlan's Databricks context layer architecture](https://atlan.com/know/context-layer-for-databricks/) makes this case platform-wide, and the underlying question, what [a data catalog needs to give an AI agent](https://atlan.com/know/data-catalog-for-ai/) rather than a human analyst, is the one this page answers for AtScale's slice. --- Inside Atlan AI Labs & the 5x Accuracy Factor See what Atlan's AI Labs measured when agents were grounded in a governed context layer instead of a single tool's metric definitions. Get the 5x Accuracy Ebook --- ## How does AtScale's MCP server use Unity Catalog's governance? AtScale shipped an MCP server, listed on the Databricks MCP Marketplace on December 1, 2025, that turns its semantic layer into something an AI agent can call directly. "Our MCP integration makes the semantic layer callable as an action service," said [Cort Johnson, SVP GTM at AtScale](https://www.atscale.com/press/atscale-databricks-mcp-marketplace-semantic-layer/). Databricks' own framing credits Unity Catalog specifically: "AtScale's MCP offering advances our shared mission and empowers organizations to deploy Agent Bricks applications," said Stephen Orban, SVP of Product Ecosystem and Partnerships at Databricks. The release states Unity Catalog's role directly: it "provides end-to-end access control and auditability," so "every agent decision is compliant, authorized, and traceable to governed business definitions." That's a real answer to a question that used to have none: does an agent calling a semantic layer inherit any governance, or is it a fresh, ungoverned surface the moment an LLM is in the loop? Inside a Databricks-only deployment, the answer for AtScale's MCP server is genuinely yes: Unity Catalog's access control, auditability, and traceability apply to the agent's call the same way they'd apply to a human analyst's query. The accurate claim is that AtScale's agent access is governed within Databricks' perimeter, not ungoverned, and overstating the gap the other way is a claim a technical buyer would catch immediately. The distinction that matters is narrower and more useful than "governed vs. ungoverned." It's governed access to one vendor's metric slice, inside one platform's boundary, versus a governed layer that compounds the same context across every platform an agent might need to reach. AtScale's own MCP launch proves the first half is real, the same access-control layer that governs agent traffic through Databricks' [Unity AI Gateway](https://atlan.com/know/ai-agent/databricks/unity-ai-gateway/). Nothing in it claims the second. AtScale's server is one entry on a fast-growing list. Agent Bricks is the Databricks surface it plugs into, [Databricks' own agent-building platform](https://atlan.com/know/ai-agent/databricks/agent-bricks/), part of [why MCP matters for AI agents](https://atlan.com/know/mcp/why-mcp-matters-for-ai-agents/) as a pattern: it standardizes how an agent asks, not whether the answer is trustworthy. That second question is the argument behind Atlan's own [MCP-connected data catalog](https://atlan.com/know/mcp-connected-data-catalog/): the protocol carries the call, not the judgment about whether it's right. --- ## What doesn't Unity Catalog give AtScale, or an agent reading through it? The plainest evidence for this boundary comes from AtScale itself: by its own description, it is "not a data catalog nor a policy engine." That's AtScale naming its own scope, and it lines up with what its Databricks integration does. Unity Catalog's own documentation backs the other half, with one caveat worth naming: its data-sharing capability lets a Databricks workspace securely share governed data and AI assets across organizations and clouds. What [Databricks' own documentation](https://docs.databricks.com/aws/en/data-governance/unity-catalog/) never claims is that Unity Catalog reaches into a second platform's own catalog and governs it on Databricks' terms. Sharing what leaves Databricks under Databricks' rules isn't the same claim as governing an estate that spans Databricks, Snowflake, and everything else. Named plainly, the gap is: no shared glossary authority between AtScale's metric definitions and anything else in the estate, no lineage surviving a second warehouse, no unstructured knowledge, and no policy layer explaining why a metric is calculated the way it is, not just what it came out to. Unity Catalog's newer [Pages feature](https://docs.databricks.com/aws/en/uc-semantics/pages) is a real attempt at closing part of that gap from the Databricks side: a Beta capability letting a team author "a governed, authoritative definition of a business concept" for Genie to cite. Worth naming honestly, but it's Beta, Databricks-only, and unconnected to AtScale's own metric layer today. Some of that gap is definitional. A metric formula, a glossary term, and an ontology relationship look alike from a distance but answer different questions, the distinction [ontology vs. semantic layer](https://atlan.com/know/ontology-vs-semantic-layer/) draws out. Neither AtScale's cube nor a populated Unity Catalog Pages entry tells an agent how to weigh either against a knowledge base it's also holding, the harder problem [agent context layer vs. knowledge base](https://atlan.com/know/ai-agent/agent-context-layer-vs-knowledge-base/) solves. On Databricks specifically, that's the gap [Genie Ontology built on Atlan's context layer](https://atlan.com/know/ai-agent/databricks/genie-ontology-and-atlan-context-layer/) closes: business meaning certified once, not authored twice and left to drift. This is where Atlan's [Context Agents and Context Engineering Studio](https://atlan.com/know/business-context-layer/) do work neither product takes on: not recomputing AtScale's metrics, but certifying the business definitions, ownership, and policy context around them, across every system an agent needs, not just the one AtScale happens to be reading from. --- AI Agent Context Readiness Checklist Check whether your agents are working from governed context or from whatever a single tool happens to expose, before you ship the next one. Get the Readiness Checklist --- ## Does the same governance gap apply if AtScale also runs on Snowflake? Briefly, yes, and the reason is structural, not incidental. [Snowflake Ventures led a strategic equity financing in AtScale, announced December 18, 2025](https://www.atscale.com/press/atscale-strategic-equity-financing-led-by-snowflake/), and since June 2026 has embedded AtScale's ACE engine directly inside its own Semantic Views: "No need for a second copy of your data and a second compute layer to process it," [AtScale wrote of the partnership](https://www.atscale.com/blog/snowflake-atscale-partnership-semantic-views/), currently in Private Preview with no public general-availability date. That's a distribution win in a Snowflake-heavy account, and a separate integration from the Databricks one this page covers, not a portable layer spanning both. AtScale treats Databricks and Snowflake as two independent, warehouse-anchored connections, so the same boundary applies twice, once per warehouse. For the full three-way decision across Snowflake Semantic Views, Cube, and AtScale, see [Snowflake Semantic Views vs Cube vs AtScale](https://atlan.com/know/ai-agent/semantic-layer/snowflake-semantic-views-vs-cube-vs-atscale/). Atlan's [Context Lakehouse](https://atlan.com/know/context-layer-vs-semantic-layer/) is the portability counter to "governed inside one warehouse's perimeter," whichever warehouse that is. The same math applies to any single-warehouse semantic layer. Snowflake's own [Semantic Views](https://atlan.com/know/snowflake/snowflake-semantic-views/) faces the identical boundary on its own platform, and so does [Cube's semantic layer](https://atlan.com/know/ai-agent/semantic-layer/cube-semantic-layer/) wherever it's deployed. --- Atlan in Action: Live Context Layer Demos Watch how Atlan governs context across a real multi-warehouse estate, not just the one platform a semantic layer happens to be embedded in. Watch the Live Demos --- ## Why a governed metric isn't a governed context layer Credit where it's real: AtScale's Unity Catalog integration is well-built, and its December 2025 MCP server genuinely answers whether an agent inherits any governance when it calls a semantic layer. Inside Databricks, yes: Unity Catalog enforces access control, keeps the audit trail, and traces a call back to AtScale's own metric definition. None of that needs softening to make the actual point. The point is narrower and holds up on its own: governed access to one vendor's metric slice, inside one platform's boundary, isn't the same claim as a layer that compounds context across every platform an agent needs. AtScale's own words settle the scope, "not a data catalog nor a policy engine", and Unity Catalog's documentation settles the rest: real controls for what leaves Databricks, no claim of governing a second platform's own catalog. Teams running AtScale alongside Atlan land on the same shape: keep AtScale for metric computation and multi-BI serving, and let a [context layer](https://atlan.com/know/data-catalog-vs-context-layer/) make those metrics, and everything else in the estate, discoverable and governed. That's coexistence, deliberately, not a consolation prize. That coexistence pattern holds regardless of which team is asking. A governance lead is really asking what [Context Layer for Data Governance Teams](https://atlan.com/know/ai-agent/context-layer-for-data-governance-teams/) answers; an engineering lead deciding [how to give AI agents access to enterprise data](https://atlan.com/know/ai-agent/how-to-give-ai-agents-access-to-enterprise-data/) safely is asking the same question from the access side. Both land on what [a semantic layer needs to give an AI agent](https://atlan.com/know/ai-agent/semantic-layer-for-ai-agents/) that a human never asked for, the shape [context engineering](https://atlan.com/know/what-is-context-engineering/) takes once an agent reads the metric. Atlan doesn't replace AtScale's metric computation, its multi-BI serving, or its aggregate-engine performance, and doesn't need to. What [Atlan's Enterprise Data Graph](https://atlan.com/know/enterprise-data-graph/) adds is the piece that stops at Databricks' edge: business definitions an agent can trust regardless of warehouse, lineage that survives crossing into a second platform, and a compounding record of what a metric means, built through [Context Engineering Studio](https://atlan.com/know/context-layer-for-ai-agents/) rather than left implicit in a cube. A governed metric answers "what's the number." An agent working across an estate eventually needs the layer that answers "can I trust where it came from," too, [worth working through before the next agent ships](https://atlan.com/know/how-to-build-ai-agent-harness/). Book a Demo --- ## FAQs about AtScale and Unity Catalog ### 1. Does AtScale support Databricks Unity Catalog? Yes. AtScale connects to Databricks over a Databricks SQL warehouse connection and imports Unity Catalog tables and views as fact and dimension datasets for its semantic models. It also reads Unity Catalog's object-level grants and relies on Unity Catalog's own row and column-level security to enforce access, rather than rebuilding an access layer of its own. ### 2. How do I connect AtScale to a Databricks SQL warehouse? You add Databricks as a data warehouse inside AtScale and point it at a Databricks SQL warehouse, then AtScale reads the relevant Unity Catalog catalogs, schemas, and tables to build fact and dimension datasets. Check AtScale's own setup documentation for the current prerequisites and permissions your Databricks admin needs to grant. ### 3. Is AtScale the same thing as Unity Catalog Metrics? No. AtScale is a third-party semantic layer vendor that predates and runs independently of Unity Catalog Metrics, Databricks' own native metrics feature announced at Data + AI Summit 2026. A team can run AtScale, Unity Catalog Metrics, both at once, or neither. ### 4. Does Unity Catalog give AtScale a business glossary, or does AtScale need one from elsewhere? Neither supplies one on its own. Unity Catalog's newer Pages feature is a real, if early, Beta capability for authoring business definitions inside Databricks, but it isn't connected to AtScale's metric definitions today. AtScale itself states it is not a data catalog or a policy engine, so a shared business glossary has to come from a layer governing both. ### 5. Can an AI agent use AtScale's metrics through Unity Catalog's governance, or are they governed separately? An agent calling AtScale's MCP server inherits Unity Catalog's access control, auditability, and traceability back to AtScale's governed metric definitions, which is real and worth crediting. What that governance doesn't extend to is the business meaning behind the metric, or anything the agent needs outside Databricks, since Unity Catalog's own documentation doesn't claim to reach into and govern a second platform's own catalog. ### 6. What if my organization runs AtScale on both Databricks and Snowflake? AtScale treats Databricks and Snowflake as two separate, warehouse-anchored integrations rather than one portable layer, so the same governance boundary applies on each side independently. For the full architecture comparison across Snowflake Semantic Views, Cube, and AtScale, see the dedicated three-way decision framework linked from this page. --- ## Sources 1. [AtScale Delivers Semantic Intelligence to Databricks MCP Marketplace, AtScale](https://www.atscale.com/press/atscale-databricks-mcp-marketplace-semantic-layer/) 2. [What is Unity Catalog?, Databricks](https://docs.databricks.com/aws/en/data-governance/unity-catalog/) 3. [Pages in Unity Catalog, Databricks](https://docs.databricks.com/aws/en/uc-semantics/pages) 4. [SML: Semantic Modeling Language, GitHub](https://github.com/semanticdatalayer/SML) 5. [Unity Catalog product page, Databricks](https://www.databricks.com/product/unity-catalog) 6. [AtScale Announces Equity Financing Led by Snowflake, AtScale](https://www.atscale.com/press/atscale-strategic-equity-financing-led-by-snowflake/) 7. [Inside the AtScale and Snowflake Partnership, AtScale](https://www.atscale.com/blog/snowflake-atscale-partnership-semantic-views/)