An AI-ready data platform gets your metric definitions and access rules to the agents that query it, for a named job. Of 12 enterprise options ranked on vendor docs, Databricks, Google Cloud, Microsoft Fabric and Snowflake pair a semantic layer with a GA agent route. Dremio and Atlan follow, and Atlan publishes one set of definitions into Snowflake and Databricks.
Gartner’s Rita Sallam, Distinguished VP Analyst, is direct about scope: “There is no way to make data AI-ready in general or in advance. The readiness of data for AI depends on how the data will be used” (Gartner). So every tool below is read for the job its own docs name, not for readiness in general.
Leaders and the people who use the data still disagree about where they stand. According to a 2026 study by COMPUTERWOCHE and Lufthansa Industry Solutions, 84% of C-level executives consider their data platforms AI-ready, while only just over half of business-area respondents agree (LHIND, 2026). Self-assessment splits by seat, so every option below is read against vendor docs instead.
- As of November 2026: all 11 platforms here document an MCP server or MCP tools, and four pair a GA route with an executable semantic layer.
- Reach is widening: Databricks, Google Cloud and Microsoft Fabric document definitions over data that other vendors store, and BigQuery made cross-cloud remote catalogs GA on November 2, 2026.
- Statuses move fast: in the first week of November 2026, Databricks put external knowledge sources in Beta and Snowflake put level of detail metrics in semantic views in Preview.
| Fact | Detail |
|---|---|
| Options ranked | 12: 11 data platforms and Atlan, a context layer across them |
| Ranking test | Executable meaning layer, then a GA agent route that uses it, then reach into other vendors’ data |
| Use case the checks were read for | Natural-language analytics over governed metrics |
| Semantic layer and GA agent route documented together | Databricks, Google Cloud, Microsoft Fabric, Snowflake |
| Evidence | Vendor product docs and release notes, opened 2026-11-06 |
What is an AI-ready data platform?
An AI-ready data platform stores and serves data and also supplies the business meaning an agent needs to use it correctly for a specific use: measures, entities, joins and the access policies that apply. Storage, compute and unstructured-data support still matter. Meaning is the axis ranked here.
The reason is mechanical. An agent that sees only schemas has to guess what “active customer” or “net revenue” means, then guess which join produces it. Pramod Sadalage, Distinguished Engineer at Thoughtworks, and Prem Chandrasekaran, Market Tech Director at Thoughtworks, put it plainly: “The semantic model doesn’t make the agent smarter. It stops it from guessing” (martinfowler.com, 2026). That gap is the difference between AI-ready and analytics-ready data: an analyst carried the meaning in their head, and an agent cannot.
Three things make a platform ready for agents:
- An executable meaning layer: semantic views, metric views, semantic models or LookML that define measures and joins an engine computes. A glossary or a tag describes data; a semantic layer computes it.
- An agent route that uses it: an MCP server or managed endpoint documented as applying that layer, with the caller’s permissions.
- Reach beyond itself: a meaning layer that can be defined over data another vendor stores.
Four of the 12 options document the first two together at GA. For the full set of checks, see what makes data AI-ready, the AI-ready data checklist and how to prepare enterprise data for AI agents.
How do the 12 options compare at a glance?
The 12 options below follow the test in the next section, read for one use case: natural-language analytics over governed metrics. Ranks 1 to 4 share the top tier and are listed alphabetically. Statuses are as each vendor’s own docs or release notes gave them on 2026-11-06.
| Rank | Platform | Executable meaning layer | Agent route that uses it (status) | Best for |
|---|---|---|---|---|
| 1 | Databricks | Unity Catalog metric views | Genie One MCP (GA), grounded in metric views | Metric definitions agents query through Genie |
| 2 | Google Cloud | LookML in Looker | Looker-managed MCP server (GA, October 2026), which queries through LookML; BigQuery MCP server (GA, April 2026) for SQL on BigQuery | LookML definitions served to agents through Looker |
| 3 | Microsoft Fabric | Power BI semantic models (DAX) | Fabric IQ MCP (GA), runs DAX against semantic models | Power BI semantic models served to agents |
| 4 | Snowflake | Semantic views (GA) | Snowflake-managed MCP server (GA), via Cortex Analyst | Semantic views served to agents through Cortex Analyst |
| 5 | Dremio | AI Semantic Layer | Dremio MCP server, hosted and self-hosted | One semantic layer over federated sources |
| 6 | Atlan | Context repository from Context Engineering Studio (Atlan Labs), rendered into Snowflake and Databricks semantic layers | Atlan MCP server; the native routes of the engines it deploys to | One set of definitions published to Snowflake and Databricks |
| 7 | AWS | DataZone business glossary; Amazon Quick Topics | AWS MCP Server (GA) for AWS APIs; Quick chat agent over Topics inside Amazon Quick | A GA agent route across AWS services |
| 8 | Oracle Autonomous AI Database | Select AI comments and annotations | Built-in MCP server (released December 2025) | Agent access to Oracle Database |
| 9 | Starburst | Data products | Starburst MCP server, read-only SQL and data product tools | Agents working through governed data products |
| 10 | Cloudera | Data Catalog glossaries | Iceberg MCP server, read-only through Impala | Read-only agent access to Iceberg through Impala |
| 11 | IBM watsonx.data | Semantic Automation Layer | watsonx.data intelligence MCP server, with Text-to-SQL | Automated enrichment in IBM Knowledge Catalog estates |
| 12 | Teradata VantageCloud | Logical data models | Teradata MCP Server, published under the MIT license | Agents on existing Teradata analytics |
Which use case is each AI-ready data tool documented for?
Every tool on this list documents its agent route for a specific job, and most document whose permissions that route runs under, so match the job before the brand. The table reads each vendor’s own docs for three things: the use case named, what carries meaning to the agent, and how access is controlled on the route.
| Platform | Use case its docs name | What carries meaning to the agent | Access on the route, as documented |
|---|---|---|---|
| Databricks | Natural-language analytics across the workspace | Metric views, synonyms, Genie Ontology (Public Preview) | Unity Catalog permissions on every request; on-behalf-of user OAuth recommended |
| Google Cloud | Agents working with business data and LookML models; conversational analytics in BigQuery | LookML | Admin tool allowlist; the agent inherits the authenticating user’s Looker roles |
| Microsoft Fabric | Finding Power BI content and running DAX | Power BI semantic models | The caller’s existing Fabric permissions |
| Snowflake | Governed business questions through Cortex Analyst | Semantic views | Permission granted per tool; sessions use the user’s default role; direct SQL on a separate server with a least-privileged role |
| Dremio | Exploring data, natural-language SQL, creating views | AI Semantic Layer | The authenticated user’s permissions |
| Atlan | One repository per use case, served to Cortex Analyst, Genie, dbt and MCP clients | Definitions built from governed catalog and glossary context, maintained by Context Agents | Atlan permissions on catalog context; each engine applies its own data permissions |
| AWS | Acting on AWS APIs; Quick chat over Topics | Business glossary terms, Quick Topics | Existing IAM credentials |
| Oracle | SQL generation with Select AI | Table and column comments, annotations | Authenticates and authorizes as the database user |
| Starburst | Searching data products and running read-only SQL | Data products with descriptions and business rules | The identity and privileges of the MCP connection |
| Cloudera | Agents on Iceberg through Impala | Data Catalog glossaries | Read-only access through Impala |
| IBM | Text-to-SQL, semantic search, glossary creation | Enrichment labels, categories and relationships | Data protection and governance tools in the MCP server |
| Teradata | Agents on Teradata analytics | Logical data models | In basic authentication mode, each request carries the user’s own database credentials |
Even one object can be ready for one job and not the next. Snowflake’s November 2, 2026 release note says level of detail (LOD) metrics “work in SQL queries against semantic views,” and that “Cortex Analyst and Cortex Agents don’t use LOD metrics yet” (Snowflake release notes). An analyst writing SQL gets the metric that week; Snowflake says its agents don’t use it yet.
For the talk-to-data case specifically, see how to prepare data for natural language analytics, the talk-to-data agent blueprint and AI-ready data for talk-to-data.
How did we rank the best AI-ready data platforms?
We ranked every option on documented capability, read from vendor product docs and release notes, opened on 2026-11-06. We read each check for one use case: an agent answering natural-language questions over governed metrics. We did not run agents against the platforms, so treat the ranking as a reading of what each vendor documents. In scope are data platforms that document both a meaning layer and an agent route over the data they host, plus one context layer that publishes definitions into them; standalone semantic-layer and catalog tools are out of scope for this list.
Check 1: Is there an executable meaning layer?
A named object that defines measures, entities or joins an engine computes counts: Databricks metric views, LookML, Power BI semantic models, Snowflake semantic views and Dremio’s semantic layer. Atlan’s context repository counts because Atlan renders it into Snowflake and Databricks formats, where those engines compute it. Glossaries, enrichment labels, annotations and data products describe data and help agents find it; they count toward Check 2, not Check 1.
Check 2: Does a GA agent route use it?
The route and the meaning layer have to be documented together: Genie One MCP grounded in metric views, the Looker MCP server working with LookML models, Fabric IQ MCP running DAX against semantic models, Cortex Analyst applying semantic views through Snowflake’s MCP server. We record a route as GA when the vendor’s docs or release notes say so (a Labs, beta or preview label is not GA). Atlan’s release-stages doc ties features enabled from Labs to private preview, so Context Engineering Studio counts as preview here.
Check 3: Can the meaning layer reach other vendors’ data?
This is the tie-breaker: is the meaning layer documented over data held in another vendor’s storage or catalog? Databricks metric views can use foreign tables from Lakehouse Federation, Looker models connect to other vendors’ warehouses, Fabric references external storage through OneLake shortcuts, Dremio federates sources such as Snowflake, and Atlan publishes one repository into Snowflake and Databricks.
How the order works
Options that pass Checks 1 and 2 at GA rank first. Next come options that document both without a GA label, with an unlabeled route ahead of one in preview. Then come platforms whose agent route works over glossaries, annotations, data products or enrichment: GA routes first, then released, then unlabeled, then preview. Within a group, Check 3 breaks ties, then the alphabet. In this edition Check 3 moved one position: among the unlabeled routes, Starburst documents reach through its Snowflake connector and ranks first, and the rest are alphabetical.
| Rank | Option | Check 1: meaning layer | Check 2: route status | Check 3: reach evidence |
|---|---|---|---|---|
| 1 | Databricks | Metric views | GA (Genie One MCP) | Foreign tables through Lakehouse Federation |
| 2 | Google Cloud | LookML | GA (Looker-managed MCP server) | Looker connections to Snowflake, Amazon Redshift, Databricks |
| 3 | Microsoft Fabric | Power BI semantic models | GA (Fabric IQ MCP) | OneLake shortcuts to Amazon S3, Google Cloud Storage, Iceberg |
| 4 | Snowflake | Semantic views | GA (managed MCP server, via Cortex Analyst) | Catalog-linked databases over remote Iceberg catalogs |
| 5 | Dremio | AI Semantic Layer | Unlabeled | Federated sources, including Snowflake |
| 6 | Atlan | Context repository, rendered for engines | Preview (Context Engineering Studio in Atlan Labs) | Deployed to Snowflake and Databricks |
| 7 | AWS | Glossary terms, Quick Topics | GA (AWS MCP Server, for AWS APIs) | DataZone third-party sources |
| 8 | Oracle | Comments, annotations | Released (December 2025) | Not scored; it changed no position |
| 9 | Starburst | Data products | Unlabeled | Snowflake connector joins across systems |
| 10 | Cloudera | Glossaries | Unlabeled | None cited here |
| 11 | IBM | Enrichment labels | Unlabeled | None cited here |
| 12 | Teradata | Logical data models | Unlabeled | None cited here |
The 12 best AI-ready data platforms at a glance
- Databricks for metric definitions agents query through Genie
- Google Cloud for LookML definitions served to agents through Looker
- Microsoft Fabric for Power BI semantic models served to agents
- Snowflake for semantic views served to agents through Cortex Analyst
- Dremio for one semantic layer over federated sources
- Atlan for one set of definitions published to Snowflake and Databricks
- AWS for a GA agent route across AWS services
- Oracle Autonomous AI Database for agent access to Oracle Database
- Starburst for agents working through governed data products
- Cloudera for read-only agent access to Iceberg through Impala
- IBM watsonx.data for automated enrichment in IBM Knowledge Catalog estates
- Teradata VantageCloud for agents on existing Teradata analytics
Best for metric definitions agents query through Genie: Databricks
Databricks keeps meaning in Unity Catalog metric views, and its Genie One MCP server is generally available. The server works through chat in Genie One, which searches dashboards, queries and metric views, so the meaning layer and the GA agent route are documented together. Metric views can also be built on foreign tables, which Lakehouse Federation connects to Snowflake, Google BigQuery, Amazon Redshift and other sources.
Pros:
- Genie One MCP is generally available, and it enforces Unity Catalog permissions on every request.
- Metric views define measures once in Unity Catalog, for SQL editors, notebooks, dashboards, Genie Agents and alerts, and for external BI tools such as Power BI, Tableau and Sigma.
- Synonyms, display names and formatting rules in metric views are built to improve agent accuracy.
- Metric views can use foreign tables, so definitions can cover data held by other vendors.
- Since October 27, 2026, ChatGPT, Codex, Claude, Claude Code and Cursor can connect to Databricks MCP servers, including Genie One MCP, with default OAuth client IDs (GA).
Good to know:
- The workspace MCP endpoints for a Genie agent, AI Search, Databricks SQL and Unity Catalog functions are in Public Preview.
- Genie Ontology, which combines metric views, domains and Pages, is in Public Preview.
- Lakehouse Federation gives read-only access to external data.
- External knowledge sources (November 5, 2026) and relationships between metric views (October 30, 2026) are both in Beta.
Databricks documents metric views as “the core implementation of Unity Catalog semantics in Unity Catalog,” with metadata to “improve agent accuracy” (Azure Databricks docs), and lists foreign tables among their supported sources (Azure Databricks docs). Its MCP docs state that “the Genie One MCP is generally available,” while the workspace endpoints “are in Public Preview” (Azure Databricks docs). The Genie One MCP server works by “leveraging Chat in Genie One,” and “Unity Catalog permissions are enforced on every request” (Azure Databricks docs). Chat in Genie One “first searches available Genie Agents for a match, then searches across dashboards, queries, and metric views,” and Genie Ontology, in Public Preview, “combines modeled context you govern in Unity Catalog semantics, such as metric views, domains, and Pages” (Azure Databricks docs), and Lakehouse Federation “provides governed, read-only access to external data through Unity Catalog foreign catalogs” (Azure Databricks docs).
With external knowledge sources, “Genie can use your Power BI measure definitions when it answers questions” (Azure Databricks release notes, November 2026), and metric view relationships record joins “so that AI assistants and BI tools choose the right join path” (Azure Databricks release notes, October 2026). More on the model: Unity Catalog metrics, Unity Catalog’s semantic layer and Genie Ontology.
| Check | What Databricks documents |
|---|---|
| Meaning layer | Unity Catalog metric views |
| Agent access | Genie One MCP, GA, through chat in Genie One; workspace endpoints in Public Preview |
| Cross-vendor scope | Metric views over foreign tables from Lakehouse Federation; Power BI semantic model metadata as an external knowledge source (Beta) |
Choose Databricks if your certified metrics live in Unity Catalog and your agents can reach them through Genie. Compare it with a cross-platform semantic layer if other platforms are in play.
Sources: Metric views · Built-in MCP services · Genie One MCP · Chat in Genie One · Lakehouse Federation · Release notes, October 2026 · Release notes, November 2026
Best for LookML definitions served to agents through Looker: Google Cloud
On Google Cloud, meaning lives in LookML, Looker’s semantic modeling language. The Looker-managed MCP server became generally available on October 26, 2026 and gives agents access to LookML models. Looker connects to a wide range of databases, including Amazon Redshift, Databricks and Snowflake, so LookML definitions can sit over other vendors’ warehouses through Looker connections. For SQL work in BigQuery itself, the BigQuery MCP server has been generally available since April 2026.
Pros:
- The Looker-managed MCP server is GA, and its query tool runs queries through the LookML model.
- The BigQuery MCP server is GA for examining resources, generating SQL and executing queries.
- LookML models can sit on Snowflake, Amazon Redshift, Databricks and other supported databases through Looker connections.
- BigQuery’s lakehouse reads Databricks Unity Catalog, AWS Glue Data Catalog and Snowflake Horizon Catalog as remote catalogs, GA since November 2, 2026.
Good to know:
- Agents reach LookML definitions through the Looker MCP server; the BigQuery MCP server serves SQL against BigQuery.
- On November 5, 2026, Google removed BigQuery’s Metadata curation page; those tasks now run in Knowledge Catalog.
- Connected agents inherit the Looker roles and access of the user who authenticated, and admins choose which tools are enabled.
LookML is “the language that is used in Looker to create semantic data models” (Google Cloud). Google’s Looker release notes for October 26, 2026 state that “the Looker-managed Model Context Protocol (MCP) server is now generally available (GA)” (Google Cloud), and the server lets agents “securely connect to a Looker instance and interact with business data and LookML models,” after which “the AI agent will inherit the Looker roles and access of the user who authenticated” (Google Cloud). Its query tool “runs an inline query using the Looker semantic model” (Google Cloud). “Looker supports a wide range of SQL database dialects,” including Snowflake, Amazon Redshift and Databricks (Google Cloud).
BigQuery’s release notes list the BigQuery MCP server, for “examining BigQuery resources” and “generating accurate and efficient SQL queries,” as “Generally Available (GA),” and on November 2, 2026 they list remote catalog providers for cross-cloud data access as GA (Google Cloud). “BigQuery Omni is a multi-cloud analytics solution” for data in Amazon S3 or Azure Blob Storage (Google Cloud). For Google’s catalog-side story, see Google Knowledge Catalog.
| Check | What Google documents |
|---|---|
| Meaning layer | LookML in Looker |
| Agent access | Looker-managed MCP server, GA; BigQuery MCP server, GA |
| Cross-vendor scope | Looker connects to Snowflake, Amazon Redshift, Databricks and other databases |
Choose Google Cloud if your metric definitions live in LookML, or your agents already run on Google Cloud.
Sources: What is LookML · Looker MCP server · Looker MCP tools · Looker release notes · BigQuery release notes
Best for Power BI semantic models served to agents: Microsoft Fabric
Microsoft Fabric pairs Power BI semantic models with Fabric IQ MCP, which is generally available and lets an agent inspect semantic models and run DAX against them with the caller’s own permissions. OneLake shortcuts reference external data in place, including Amazon S3, Google Cloud Storage and Iceberg sources.
Pros:
- Fabric IQ MCP is generally available.
- MCP calls apply the caller’s existing Fabric permissions.
- Agents run DAX queries against Power BI semantic models, using the measures defined there.
- Semantic models in Direct Lake mode can read shortcut data directly.
Good to know:
- The Fabric IQ ontology is in preview; Fabric IQ MCP works with Power BI semantic models.
- Fabric IQ MCP is a read-only toolset.
Microsoft Learn describes Fabric IQ’s two core items as “ontology (preview) and semantic model” and says Fabric IQ can “unify external operational data using OneLake shortcuts, referencing it in place without copying” (Microsoft Learn). It states that “Fabric IQ MCP is generally available,” that the server “exposes a read-only toolset that helps an agent find Power BI content, inspect report and semantic model metadata, locate relevant values, and run DAX queries,” and that it “applies the caller’s existing Microsoft Fabric permissions” (Microsoft Learn). With Direct Lake, “Analysis Services can read data directly from the shortcut” (Microsoft Learn). For how Fabric IQ relates to Microsoft’s other context sources, see Fabric IQ vs Work IQ vs Foundry IQ, and for the other servers in the suite, what Fabric’s own MCP servers let an agent do.
| Check | What Microsoft documents |
|---|---|
| Meaning layer | Power BI semantic models (DAX); Fabric IQ ontology in preview |
| Agent access | Fabric IQ MCP, GA, read-only, runs DAX with the caller’s permissions |
| Cross-vendor scope | OneLake shortcuts to Amazon S3, Google Cloud Storage, Iceberg and more |
Choose Fabric if Power BI semantic models already hold your metric definitions.
Sources: Fabric IQ overview · Fabric IQ MCP · OneLake shortcuts
Best for semantic views served to agents through Cortex Analyst: Snowflake
Snowflake stores business meaning in semantic views, schema-level objects that are generally available, with Semantic View Autopilot to help create them. Its Snowflake-managed MCP server, generally available since November 4, 2025, serves Cortex Analyst, Cortex Search and Cortex Agents as tools, and applies semantic views through Cortex Analyst.
Pros:
- Semantic views live in the database as schema-level objects.
- Semantic View Autopilot creates semantic views with AI assistance; views are then refined and deployed in Semantic Studio, and the docs cover Tableau and Power BI ingestion.
- One managed MCP server can serve Cortex Analyst, Cortex Search and Cortex Agents.
Good to know:
- Snowflake applies semantic views through the Cortex Analyst tool, and recommends exposing direct SQL on a separate MCP server so governed questions go through the semantic views.
- Level of detail (LOD) metrics in semantic views are in Preview (November 2, 2026), and Snowflake’s release note says “Cortex Analyst and Cortex Agents don’t use LOD metrics yet.”
Snowflake’s docs say “you can store semantic business concepts directly in the database in a Semantic View, which is a schema-level object” (Snowflake docs), and its release notes state that “the ability to query semantic views is now generally available” (Snowflake release notes). “Semantic View Autopilot is the AI-assisted way to create a semantic view in Snowsight,” and “Autopilot covers creation. After the view exists, you refine and deploy it in Semantic Studio” (Snowflake docs).
The MCP docs say “you can configure the MCP server to serve Cortex Analyst, Cortex Search, and Cortex Agents as tools” and that the server “only supports using semantic views with Cortex Analyst.” They also warn that “exposing SYSTEM_EXECUTE_SQL on the same server allows the MCP client to bypass the agent’s semantic views, verified queries, and orchestration” (Snowflake docs). Sessions use “the connecting user’s DEFAULT_ROLE as the primary role,” and “permission needs to be granted for each tool.” The November 4, 2025 release note announced “the general availability of the Snowflake-managed MCP server” (Snowflake release notes). For data catalogued elsewhere, a catalog-linked database “automatically discovers and stays in sync with the namespaces and tables in your remote catalog” (Snowflake docs). Go deeper with Snowflake semantic views and Cortex Analyst vs custom text-to-SQL.
| Check | What Snowflake documents |
|---|---|
| Meaning layer | Semantic views (GA), Semantic View Autopilot; LOD metrics in Preview |
| Agent access | Snowflake-managed MCP server, GA since November 4, 2025, via Cortex Analyst |
| Cross-vendor scope | Catalog-linked databases that sync tables from remote Iceberg REST catalogs, such as AWS Glue |
Choose Snowflake if your agents’ questions resolve to Snowflake data and Cortex Analyst can be their route to the definitions. For context across more systems, see a context layer for Snowflake.
Sources: Semantic views overview · Semantic View Autopilot · Snowflake-managed MCP server · MCP server GA release note · LOD metrics release note · Apache Iceberg tables
Best for one semantic layer over federated sources: Dremio
Dremio provides a semantic layer with shared definitions for business entities, metrics and relationships, over live access to lakes, warehouses and databases without ETL pipelines or copies. Its MCP server, available Dremio-hosted or self-hosted, lets agents explore data, generate SQL from natural language and create views, so the semantic layer and the agent route are documented together.
Pros:
- One semantic layer over live access to lakes, warehouses and databases, which Dremio documents as needing no ETL pipelines or copies.
- Snowflake is a documented source, alongside catalogs such as AWS Glue and Unity Catalog.
- Hosted MCP for Claude.ai and ChatGPT; self-hosted MCP for Claude Code, Cursor and other local clients.
- Agents work with Dremio’s existing access controls.
Good to know:
- Choose the hosted or self-hosted MCP option based on which AI clients your team uses.
Dremio’s docs say “the semantic layer provides shared definitions for business entities, metrics, and relationships” and that Dremio Cloud “provides live access to data across lakes, warehouses, and databases without ETL pipelines or copies,” with “integration with catalogs such as AWS Glue and Unity Catalog” (Dremio docs). Its MCP server can “explore data, generate SQL queries from natural language, and create views in Dremio,” and a tool “retrieves descriptions and tags for a table or schema” (Dremio docs). On the hosted and self-hosted options, “agents operate with the same permissions as the authenticated user” (Dremio docs), and Snowflake is a supported source (Dremio docs).
| Check | What Dremio documents |
|---|---|
| Meaning layer | AI Semantic Layer |
| Agent access | Dremio MCP server, hosted and self-hosted |
| Cross-vendor scope | Federated sources, including Snowflake, AWS Glue and Unity Catalog |
Choose Dremio if your data already sits in several stores and you want one semantic layer over them without an ingestion project.
Sources: Dremio Cloud overview · Dremio MCP server · MCP server tools
Best for one set of definitions published to Snowflake and Databricks: Atlan
Atlan’s Context Engineering Studio assembles a context repository for one use case from governed catalog and glossary context, after a person approves the assets it proposes, and renders it into the semantic layer each engine expects: a semantic view for Snowflake Cortex Analyst, metric views and a Genie space for Databricks, MetricFlow models for dbt. MCP-capable agents such as Claude read the same repository through the Atlan MCP server. Atlan works as a context layer across these platforms rather than replacing one.
Pros:
- One repository renders into Snowflake semantic views, Databricks metric views and dbt MetricFlow models.
- Snowflake and Databricks deploy directly from Context Engineering Studio, so agents query the warehouse natively.
- Claude, OpenAI, Gemini CLI and other MCP clients read the repository through the Atlan MCP server.
- Context Agents generate and maintain descriptions, metrics and definitions across the estate, and a person approves the assets each repository uses.
Good to know:
- Context Engineering Studio is enabled per tenant in Atlan Labs, which Atlan’s release-stages doc ties to private preview; your Atlan account team enables it.
- The Atlan MCP server applies Atlan permissions to catalog context; each engine runs its own queries under its own data permissions.
Atlan’s docs say to “describe the use case, approve the assets CES proposes, and you get a context repository,” built “from the governed context in your Atlan catalog and glossary,” then rendering it “as a semantic view for Snowflake Cortex Analyst, metric views for Databricks Genie, MetricFlow models for dbt, and context Claude reads over MCP,” with Context Agents that “generate and maintain descriptions, metrics, and definitions across your estate” (Atlan docs). The Databricks target “publishes the repository as Databricks metric views and a Genie space, so agents query the warehouse natively,” and the MCP target lets “Claude, OpenAI, Gemini CLI, and other MCP clients” read the repository directly (Atlan docs). Studio is enabled “on your Atlan tenant in Atlan Labs” (Atlan docs), and Atlan’s release-stages page says a private preview feature “requires admins to enable the feature from Labs in the admin center” (Atlan docs).
The Atlan MCP server itself is “a hosted server, available on every Atlan tenant with nothing to install,” and “every action respects the permissions you’ve already set in Atlan,” with tools covering “search, lineage, metadata, glossary, governance, and data quality” (Atlan docs). For how the server works, see what is Atlan MCP.
| Check | What Atlan documents |
|---|---|
| Meaning layer | Context repository: dimensions, facts, relationships and derived metrics, rendered into each engine’s semantic format |
| Agent access | Atlan MCP server (hosted on every tenant); native routes of the engines it deploys to; Context Engineering Studio in Atlan Labs (private preview) |
| Cross-vendor scope | One repository deployed to Snowflake and Databricks and rendered for dbt |
Choose Atlan if your agents work across Snowflake and Databricks, or read context through MCP, and you want one set of governed definitions published to each. The trade-offs are covered in single-stack lock-in vs a neutral context layer, the multi-cloud context layer and the Atlan context layer as enterprise memory.
Sources: What is Context Engineering Studio · Portability targets · Atlan MCP overview · Product release stages
Best for a GA agent route across AWS services: AWS
AWS gives agents a generally available AWS MCP Server that can call any AWS API through a single tool, with API calls authenticated through existing IAM credentials. Meaning sits in two other places: business terms in the Amazon DataZone business glossary, which covers data stored across AWS, on premises and third-party sources, and Amazon Quick Topics, the multi-dataset semantic layer the Quick chat agent uses inside Amazon Quick.
Pros:
- AWS MCP Server is generally available.
- One tool calls any AWS API, authenticated with your existing IAM credentials.
- DataZone catalogs and governs data across AWS, on-premises and third-party sources.
- Quick Topics hold dataset relationships and custom instructions that the Quick chat agent uses to write cross-dataset SQL.
Good to know:
- A Topic holds up to 12 datasets, and Topics apply to Amazon Quick Enterprise Edition.
AWS announced “the general availability of the AWS MCP Server” on May 6, 2026, adding that “agents can now call any AWS API through a single tool” (AWS What’s New). To execute AWS API calls, “agents authenticate through your existing IAM credentials” (AWS docs). DataZone is a service to “catalog, discover, share, and govern data stored across AWS, on premises, and third-party sources” (AWS docs), and “a business glossary is a collection of business terms (words) that may be associated with assets” (AWS docs). In Amazon Quick, “a Topic in Quick Sight is the multi-dataset semantic layer that brings multiple enriched datasets together into a unified data model” (AWS docs), and the chat agent “uses your defined relationships, dataset enrichment metadata, and custom instructions to generate cross-dataset SQL queries” (AWS docs). For how DataZone and Glue divide the work, see AWS DataZone vs Glue Data Catalog and the Amazon DataZone data catalog.
| Check | What AWS documents |
|---|---|
| Meaning layer | DataZone business glossary; Amazon Quick Topics |
| Agent access | AWS MCP Server, GA, for AWS APIs; Quick chat agent inside Amazon Quick |
| Cross-vendor scope | DataZone catalogs on-premises and third-party sources |
Choose AWS if your agents mainly act on AWS services. For what crosses the AWS boundary, read AWS Glue and DataZone vs a cross-cloud context layer.
Sources: AWS MCP Server GA · AWS MCP Server docs · DataZone concepts · DataZone business glossary · Quick Topics · Topics in Quick chat
Best for agent access to Oracle Database: Oracle Autonomous AI Database
Oracle Autonomous AI Database includes a built-in, managed MCP server, released in December 2025, that exposes Select AI Agent tools and custom tools to outside AI applications. Select AI can draw on table and column comments and annotations stored in the database when it generates SQL.
Pros:
- Managed, multi-tenant MCP server, so no server to run yourself.
- Available for Autonomous AI Database 26ai and 19c on Serverless Exadata Infrastructure.
- Select AI Agent tools and custom MCP tools for common database operations.
- Select AI uses comments and annotations as metadata when generating SQL.
Good to know:
- The MCP server uses the database user for authentication and authorization.
Oracle’s release notes, dated December 23, 2025, say “Autonomous AI Database now includes a built-in MCP Server that provides a managed, per-database MCP server,” for Autonomous AI Database 26ai and 19c on Serverless Exadata Infrastructure (Oracle release notes). Its documentation describes “a managed, multi-tenant server designed to provide secure, standardized access to database tools and features through the Model Context Protocol” (Oracle docs), and “the Autonomous AI Database MCP Server uses the database user for authentication and authorization” (Oracle docs). AI profiles can “use the table and column comments from the data dictionary to better understand the data” (Oracle docs).
| Check | What Oracle documents |
|---|---|
| Meaning layer | Select AI comments and annotations |
| Agent access | Built-in MCP server, released December 23, 2025 |
Choose Oracle if Oracle Database holds your system of record and agents need governed access to it.
Sources: MCP server release note · Autonomous AI Database MCP server · MCP server concepts · AI profiles
Best for agents working through governed data products: Starburst
Starburst Enterprise includes an integrated MCP server that lets agents run read-only SQL and search and retrieve data products, and Starburst Galaxy offers MCP as well. Starburst documents data products as a semantic layer over underlying sources, each with at least one named owner, and its Snowflake connector lets queries join Snowflake data with other systems.
Pros:
- Integrated MCP server in Starburst Enterprise, plus MCP on Starburst Galaxy.
- Agents can search data products and retrieve their details, including business rules, through the searchDataProducts and getDataProductDetails MCP tools.
- Every published data product has at least one named owner.
- The Snowflake connector joins Snowflake data with other systems in one query.
Good to know:
- The MCP query tool runs read-only SQL.
- On Starburst Enterprise, the agent runs with the identity configured in the MCP connection.
Starburst’s docs state that “Starburst Enterprise includes an integrated Model Context Protocol (MCP) server,” that “the query tool supports only read-only SQL queries,” and that data products are searchable through the searchDataProducts tool, while “the agent runs with the identity configured in the MCP connection” (Starburst docs). Galaxy’s MCP server lets AI tools “execute read-only SQL queries against your data through natural language interactions” (Starburst docs).
The docs tell teams to “use data products as a semantic layer to abstract the technical complexity of your underlying data sources” (Starburst docs), and publishing one requires “at least one data product owner,” with an optional business rules field that gives Starburst’s AI agent context such as “how a metric is calculated” (Starburst docs). The Snowflake connector “can be used to join data between different systems like Snowflake and Hive” (Starburst docs).
| Check | What Starburst documents |
|---|---|
| Meaning layer | Data products with descriptions, business rules and owners |
| Agent access | Integrated MCP server (Enterprise), MCP on Galaxy; read-only SQL and data product tools |
Choose Starburst if you want agents to find and query data through governed data products with named owners.
Sources: Starburst MCP server · MCP on Starburst Galaxy · Data products overview · Snowflake connector
Best for read-only agent access to Iceberg through Impala: Cloudera
Cloudera publishes an Iceberg MCP server that gives agents read-only access to Apache Iceberg tables through Apache Impala, and Cloudera Data Catalog glossaries hold business terms.
Pros:
- Built for Apache Iceberg tables, queried through Impala.
- Read-only access, so agents query tables without changing them.
- The server’s code is public on GitHub.
- Data Catalog glossaries define a hierarchical set of business terms.
Good to know:
- Check current support status for the Iceberg MCP server with Cloudera before production use.
Cloudera’s Iceberg MCP server “provides read-only access to Iceberg tables via Apache Impala” (Cloudera Iceberg MCP Server). “Glossaries enable you to define a hierarchical set of business terms that represents your business domain” (Cloudera docs).
| Check | What Cloudera documents |
|---|---|
| Meaning layer | Cloudera Data Catalog glossaries |
| Agent access | Iceberg MCP server, read-only through Impala |
Choose Cloudera if your Iceberg tables are served through Impala and agents need read-only access to them.
Sources: Cloudera Iceberg MCP Server · Data Catalog glossaries
Best for automated enrichment in IBM Knowledge Catalog estates: IBM watsonx.data
IBM watsonx.data uses the Semantic Automation Layer in IBM Knowledge Catalog to add labels, categories and relationships with AI. The watsonx.data intelligence MCP server gives agents tools across governance, lineage, data products, semantic search and query, and Text-to-SQL.
Pros:
- Semantic automation adds labels, categories and relationships to data.
- The MCP server covers Text-to-SQL, semantic search and query, lineage and data products.
- An MCP tool builds glossaries from business documents.
Good to know:
- Semantic enrichment requires registering IBM Knowledge Catalog through the watsonx.data web console.
IBM’s docs say “semantic automation for data enrichment is a new feature introduced in IBM watsonx.data,” which “leverages Semantic Automation Layer (SAL) in IBM Knowledge Catalog” (IBM docs). The watsonx.data intelligence MCP server lists tool areas including “Semantic search and query” and “Text-to-SQL” (IBM docs), and its create_glossary_from_files tool “extracts business terms from these documents” (IBM docs). To turn it on, “you must register IBM Knowledge Catalog through the web console” (IBM docs).
| Check | What IBM documents |
|---|---|
| Meaning layer | Semantic Automation Layer (labels, categories, relationships) |
| Agent access | watsonx.data intelligence MCP server, with Text-to-SQL and semantic search tools |
Choose watsonx.data if IBM Knowledge Catalog already governs your data and you want enrichment automated.
Sources: Semantic automation · watsonx.data intelligence MCP server · Glossaries via MCP
Best for agents on existing Teradata analytics: Teradata VantageCloud
Teradata publishes the Teradata MCP Server under the MIT license to “connect AI agents directly to Teradata with enterprise security and extensibility.” In basic authentication mode, every request carries the user’s own database credentials. Teradata also documents logical data modeling for its database.
Pros:
- The server’s code and documentation are public, under the MIT license.
- Basic authentication mode requires valid database credentials with each request.
- Teams can add their own tools and business logic to the server.
Good to know:
- With authentication set to none, all requests use the server’s database credentials; choose basic mode when access should follow each user.
Teradata’s repository describes the server as a way to “connect AI agents directly to Teradata with enterprise security and extensibility” (Teradata MCP Server). Its configuration guide says that in basic mode “users must provide valid database credentials with each request,” while with no authentication “all requests use the server’s database credentials” (Teradata MCP Server docs), and its docs invite teams to “add your own tools and business logic” (Teradata MCP Server docs). Its database design documentation “describes the basics of semantic, or logical, data modeling” (Teradata docs).
| Check | What Teradata documents |
|---|---|
| Meaning layer | Logical data models |
| Agent access | Teradata MCP Server; basic mode uses each user’s database credentials |
Choose Teradata if your analytics run on Teradata today and you want agents on that data without a migration.
Sources: Teradata MCP Server · Server configuration · Server guide · Semantic data modeling
What are the other ways to give agents shared definitions?
There are two routes beside the 12 above. An agent can call several platforms’ MCP servers directly, Databricks for one question and Snowflake for the next, with “revenue” defined on each platform. Or a team can define metrics once in a headless semantic layer: the dbt Semantic Layer lets teams “define metrics on top of existing models” and handles the joins, and AI tools “can connect to the Semantic Layer through the dbt MCP server” (dbt docs).
Other headless layers are compared in Snowflake semantic views vs Cube vs AtScale. A context layer keeps one set of owned definitions in sync across engines; see full-stack AI platform vs best-of-breed context layer and context layer total cost of ownership.
How do enterprises choose the right AI-ready data platform?
Enterprises choose by where certified definitions live today, which use case the first agent serves, and how the agent route controls access. Start with four questions: which platform already holds your metric definitions, which one use case ships first, how many platforms an agent needs per question, and who owns and certifies definitions.
| If your estate looks like… | Start with… | Why |
|---|---|---|
| One platform, no cross-platform joins | That platform’s native meaning layer | Native context may be enough; don’t overbuy |
| Certified metrics in Unity Catalog | Databricks | Genie One MCP is GA and grounded in metric views |
| Metrics modeled in LookML | Google Cloud | The Looker-managed MCP server is GA and works with LookML models |
| Metrics already modeled in Power BI | Microsoft Fabric | Fabric IQ MCP is GA and runs DAX against semantic models |
| Semantic views in Snowflake | Snowflake | The managed MCP server is GA and applies semantic views through Cortex Analyst |
| Agents answer from Power BI measures but run on Databricks | Databricks external knowledge sources, or a context layer | Definitions stay in Power BI; the Databricks route is in Beta |
| Two or more warehouses, data must stay put | Dremio, or a headless semantic layer | One semantic layer over federated sources |
| Access rules must hold on the agent route | A route that runs as the person asking, such as Fabric IQ MCP, Genie One MCP or Dremio | Data access follows the caller’s own permissions |
| Agents span Snowflake, Databricks and dbt, or read context over MCP | Atlan | One context repository rendered into each of their semantic layers |
One platform: use what ships with it, and check that the agent route you need is GA and applies the semantic layer. If your catalog already holds descriptions and owners, extend the data catalog for AI readiness before you buy anything new.
Two platforms: decide where definitions are authored: in each platform, in a federation engine, in a headless semantic layer or in a context layer. The point solution vs context layer comparison lays out the trade-off, and an AI readiness assessment shows which gaps sit in your data rather than in the platform.
Estate-wide agents: plan for definitions with an owner and a review step. Teams weighing a homegrown route should read building a DIY context layer and how to build MCP servers for enterprise data, and how to implement an enterprise context layer for AI covers the rollout.
Test the shortlist on your own questions. Pick 20 to 30 business questions with certified answers and run each through the exact agent route you plan to ship. Log the route, the SQL or DAX it generated and the identity it ran as. Score answers against the certified ones, count the questions the agent should have declined, and run the set again after a schema change.
Why do AI-ready platforms still give agents wrong answers?
Agents on AI-ready platforms still answer wrongly when the agent route skips the meaning layer, when definitions drift after a schema change, or when meaning covers only part of the question. Platform features give agents a route; keeping definitions correct still needs an owner.
Watch for four mistakes:
- Pointing the agent at the generic route. Snowflake’s docs warn that exposing direct SQL on the same MCP server “allows the MCP client to bypass the agent’s semantic views, verified queries, and orchestration,” and recommend a separate server for it. See why text-to-SQL fails without a context layer and text-to-SQL vs a semantic layer for self-serve analytics.
- Modeling meaning in one platform while agents also query another through its own route. The second route never sees the first platform’s definitions.
- Shipping definitions with no owner. Schema changes break them silently, and data quality for AI degrades with them. Owners and stewards for AI-ready data certify fitness for each use.
- Calling data AI-ready without naming the use case. A metric can be ready for an analyst’s SQL and not for the agent route in the same week.
Real stories from real customers: shared language for AI
"Atlan captures Workday's shared language to be leveraged by AI via its MCP server. As part of Atlan's AI labs, we're co-building the semantic layer that AI needs."
Joe DosSantos, VP, Enterprise Data and Analytics, Workday
Where should definitions live once agents span platforms?
Four platforms on this list document a semantic layer and a GA agent route that applies it: Databricks, Google Cloud, Microsoft Fabric and Snowflake. Databricks, Google Cloud and Fabric document definitions over data that other vendors store, and Snowflake syncs tables from remote Iceberg catalogs.
So the platform choice and the meaning choice are separate decisions. Pick the platform for where your data lives. Then decide where definitions are authored and owned: in one platform, in a headless semantic layer, or in a context layer that publishes to all of them.
Gartner “advises data and analytics (D&A) leaders to establish a context layer as a core component of their D&A infrastructure” (Gartner, 2026). The enterprise context layer explains that option, what is context engineering covers the practice behind it, and data infrastructure for AI covers the rest of the stack. Name the first use case, then check which of the 12 routes serves it today.
FAQs about AI-ready data platforms
1. What is the best AI-ready data platform?
There is no single best platform. Read for natural-language analytics over governed metrics, Databricks, Google Cloud, Microsoft Fabric and Snowflake rank highest: each documents a semantic layer and a GA agent route that applies it. The right one is usually the platform that already holds your certified metric definitions.
2. What are the best AI-ready data tools for enterprises?
The same four lead, listed alphabetically, followed by Dremio and then Atlan, which publishes one set of definitions into Snowflake and Databricks. For enterprises running several platforms, the deciding questions are where governed definitions are authored and which use case ships first.
3. What makes a data platform AI-ready?
A data platform is AI-ready for a named use case when it has a semantic layer that defines measures and joins, an agent route such as an MCP server that applies that layer, and access policies on that route. Storage, compute and vector search alone leave an agent guessing at definitions.
4. Which data platforms have an MCP server?
All 11 platforms here document MCP servers or MCP tools, and so does Atlan. Databricks Genie One MCP, the Looker-managed and BigQuery MCP servers, Fabric IQ MCP, the Snowflake-managed MCP server and the AWS MCP Server are GA. Oracle released its built-in MCP server in December 2025.
5. Is Snowflake or Databricks more AI-ready?
On documented capability they are close. Both keep executable meaning natively, in semantic views and metric views, and both have a GA agent route that uses it: Cortex Analyst through Snowflake’s managed MCP server, and Genie One MCP. Databricks also documents metric views over foreign tables, and its external knowledge sources, in Beta since November 2026, let Genie use Power BI measure definitions. Snowflake documents catalog-linked databases that sync tables from remote Iceberg catalogs such as AWS Glue.
6. Do I need a semantic layer if my warehouse has AI features?
For open-ended metric questions, usually yes. AI features let an agent query the warehouse, but without a semantic layer it guesses what metrics mean and how tables join. A semantic layer hands the agent computed definitions instead of column names to interpret. For narrow tasks, approved queries or domain APIs can give consistent answers without one.
7. Which AI-ready data platform has built-in governance?
Several document access control on the agent route itself. Fabric IQ MCP applies the caller’s existing Fabric permissions, Dremio agents work with the same permissions as the authenticated user, Genie One MCP enforces Unity Catalog permissions on every request, and every Atlan MCP action respects the permissions already set in Atlan. Check the route your agent will use, not the platform in general.
8. Can one platform’s semantic layer describe data in another platform?
Yes, in several ways. Databricks metric views can use foreign tables, Looker connects LookML to Snowflake, Amazon Redshift and Databricks, Fabric references external storage through OneLake shortcuts, and Dremio federates sources such as Snowflake. Atlan renders one context repository into Snowflake and Databricks semantic layers.
9. What is the first step to making data AI-ready?
Pick one business question your agents must answer and trace it end to end: the data it needs, the definitions it depends on, and who owns them. That exposes missing meaning faster than a platform evaluation.
10. How do I check whether a vendor’s agent features are GA?
Read the vendor’s current documentation and release notes for an explicit status label, and trust the dated release note when a docs page lags behind it. A Labs, beta or preview label is not GA. Recheck the route you plan to ship before you commit.
Sources
- AI-Ready Data Essentials to Capture AI Value, Gartner
- Study: AI-ready data platforms 2026, Lufthansa Industry Solutions
- Making data ready for agentic AI, martinfowler.com
- Level of detail (LOD) expressions in semantic views, Snowflake release notes
- Unity Catalog metric views, Microsoft Learn (Azure Databricks)
- Metric view modeling, Microsoft Learn (Azure Databricks)
- Built-in MCP services, Microsoft Learn (Azure Databricks)
- Genie One, Microsoft Learn (Azure Databricks)
- Lakehouse Federation, Microsoft Learn (Azure Databricks)
- Azure Databricks release notes, November 2026, Microsoft Learn
- Azure Databricks release notes, October 2026, Microsoft Learn
- What is LookML, Google Cloud
- Looker release notes, Google Cloud
- Looker MCP server, Google Cloud
- Looker dialect support, Google Cloud
- BigQuery release notes, Google Cloud
- Introduction to BigQuery Omni, Google Cloud
- Fabric IQ overview, Microsoft Learn
- Fabric IQ MCP, Microsoft Learn
- OneLake shortcuts, Microsoft Learn
- Overview of semantic views, Snowflake docs
- Release notes: semantic views generally available, Snowflake
- Semantic View Autopilot, Snowflake docs
- Snowflake-managed MCP server, Snowflake docs
- Snowflake-managed MCP server general availability, Snowflake release notes
- Dremio Cloud overview, Dremio docs
- Dremio MCP server tools, Dremio docs
- Dremio MCP server, Dremio docs
- Connect to Snowflake, Dremio docs
- What is Context Engineering Studio, Atlan Documentation
- Context Engineering Studio portability targets, Atlan Documentation
- Enable Context Engineering Studio, Atlan Documentation
- Atlan MCP overview, Atlan Documentation
- The AWS MCP Server is now generally available, AWS What’s New
- AWS MCP Server, Agent Toolkit for AWS docs
- Amazon DataZone terminology and concepts, AWS
- Create and maintain a business glossary, AWS
- Working with Amazon Quick Sight Topics, AWS docs
- Using Topics in Amazon Quick chat, AWS docs
- MCP Server for Autonomous AI Database release note, Oracle
- Autonomous AI Database MCP server, Oracle docs
- Create an AI profile, Oracle docs
- Starburst MCP server, Starburst docs
- MCP on Starburst Galaxy, Starburst docs
- Data products overview, Starburst docs
- Manage data products, Starburst docs
- Starburst Snowflake connector, Starburst docs
- Cloudera Iceberg MCP Server, Cloudera (GitHub)
- Data Catalog glossaries, Cloudera docs
- Semantic automation for data enrichment, IBM docs
- watsonx.data intelligence MCP server, IBM docs
- Creating glossaries with MCP, IBM docs
- Teradata MCP Server, Teradata (GitHub)
- Teradata MCP Server configuration guide, Teradata (GitHub)
- Teradata MCP Server documentation, Teradata (GitHub)
- Semantic data modeling, Teradata docs
- dbt Semantic Layer, dbt Labs
- Gartner Says Lack of Semantics Causes Inaccurate AI Agents and Wasted Spending, Gartner
- Genie One MCP server, Microsoft Learn (Azure Databricks)
- McpTools, Looker API reference, Google Cloud
- Apache Iceberg tables, Snowflake docs
- Autonomous AI Database MCP server concepts, Oracle docs
- Product release stages, Atlan Documentation