--- url: "https://www.xcademia.com/news/google-cloud-s-data-agent-kit-brings-agentic-analytics-into-the-ide" title: "Google Cloud's Data Agent Kit Brings Agentic Analytics Into the IDE" description: "Google Cloud's Data Agent Kit connects agents to BigQuery, Cloud SQL and Cloud Storage for agentic analytics, root cause analysis and data workflows." publishedAt: "2026-09-09T07:06:42.702+00:00" updatedAt: "2026-09-09T08:33:23.919031+00:00" type: news category: "ai-ml" source_name: Google Cloud Blog source_url: "https://cloud.google.com/blog/products/data-analytics/agentic-analytics-with-the-data-agent-kit" tags: - "#DataAnalytics" - "#AgenticAI" - "#GoogleCloud" - "#DataAgents" - "#BigQuery" - "#ModelContextProtocol" - "#MCP" - "#AIEngineering" --- # Google Cloud's Data Agent Kit Brings Agentic Analytics Into the IDE > Google Cloud's Data Agent Kit lets data teams investigate business questions across BigQuery, Cloud SQL and Cloud Storage from an agentic IDE, turning open-ended analysis into a guided, reproducible workflow. Source: **Google Cloud Blog** · 9 September 2026 ## Google Cloud is bringing agentic analytics closer to the data developer's workflow A business question can sound simple while requiring a surprisingly complex investigation. Consider a director asking: average order value dropped in January, but total revenue remained flat. What happened? For a data practitioner, answering that question may require more than a single SQL query. Order history could be stored in a data warehouse, customer information could live in a PostgreSQL database, while campaign rules may exist as JSON files in object storage. The analyst has to move between systems, understand different schemas and query dialects, connect the results and determine which pieces actually explain the change. Google Cloud's newly highlighted [**Data Agent Kit**](https://docs.cloud.google.com/data-agent-kit) is designed to reduce that friction. The toolkit combines Model Context Protocol, or MCP, servers with agent skills so data developers can work with data systems directly from supported agentic development environments. Instead of simply generating SQL for a developer to copy into another console, the agent can invoke tools, execute queries and inspect results as part of the investigation. The Data Agent Kit is available as an extension for VS Code forks including Antigravity IDE and Cursor, and as a [plugin](https://github.com/gemini-cli-extensions/data-agent-kit-starter-pack)for tools including Antigravity 2.0, Antigravity CLI, Claude Code and Codex. Google says the kit is currently in preview. ### ![info-1](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1788936826764-info-1--136---1-.webp) ## What is the Data Agent Kit? Google describes the [Data Agent Kit](https://docs.cloud.google.com/data-agent-kit) as a collection of **MCP servers and agent skills** intended to help data developers run data workflows from their IDEs. Two mechanisms are central to the approach. **Model Context Protocol** MCP is used as the connection layer between an agent and external tools, databases and cloud infrastructure. In the example provided by Google Cloud, the agent uses MCP connections to move between [BigQuery](https://cloud.google.com/bigquery), [Cloud SQL](https://cloud.google.com/sql) and [Cloud Storage](https://cloud.google.com/storage) as the investigation progresses. **Agent skills** Skills are Markdown files that add knowledge to the agent and teach it how to interact with a particular data stack. Together, these mechanisms allow the agent to do more than produce a suggested query. It can use the available tools, retrieve results and continue an investigation based on what it finds. The distinction matters because many data investigations are iterative. The first query rarely provides the complete explanation. ## From an open-ended question to a measurable problem Google's example begins with an average order value investigation. The analyst first asks the agent to calculate monthly average order value from August 2025 through January 2026 using order and order-item tables in BigQuery. Before executing the required MCP tools, the IDE requests permission. Google notes that users can approve access for auditing or choose an "always allow" option to continue the workflow. The agent then sends the relevant queries to BigQuery. Importantly, the workflow remains inspectable. The execution trail can show MCP tool calls and the raw SQL sent to BigQuery. That provides an opportunity for the developer to review what the agent actually did rather than treating the result as a black box. **The first finding** The example shows average order value staying around **$110 from August through December**, before falling to **$103 in January**. Rather than stopping at that observation, the user asks the agent to break January's AOV down by order type. That additional analysis changes the interpretation. Online and Offline orders remained around $110. Instead, a new **B2B-Wholesale** channel appeared in January with an AOV of approximately $75. The issue was therefore not that existing channels suddenly became weaker. The blended average was affected by a change in product and order mix. ### ![info-2](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1788936846515-info-2--117---1-.webp) ## The investigation continues into Cloud SQL The next question is no longer about the metric itself. The analyst wants to understand who the new wholesale buyers are. Those records are stored in a Cloud SQL PostgreSQL operational database. Because the investigation remains in the same chat session, the user can ask the agent to inspect the Cloud SQL customer table. Google's example says the agent finds **100 wholesale accounts**, all of them new business entities created within the previous 30 days. None existed in December. This demonstrates one of the central ideas behind the Data Agent Kit: the investigation does not have to stop when the required information moves into another system. The agent can change tools as the question changes. ## From customer records to campaign files The investigation then moves back to BigQuery. The example shows that **92% of the B2B orders used the **`BIGORDER25`** promo code**. The next step is to identify what that code actually represented. Instead of manually locating the campaign information, the agent uses the Google Cloud Storage MCP server to access the campaign file. The marketing campaign reveals that the code provided a **25% discount**, resulting in a large number of lower-priced wholesale orders. That explains how the blended average order value fell while total revenue remained flat. The root cause was distributed across three different data environments: - **BigQuery:** identified the AOV change and order mix - **Cloud SQL:** identified the new wholesale customer accounts - **Cloud Storage:** revealed the campaign and discount behind the orders Google's example demonstrates how the agent can connect those pieces within one investigation. ## Turning analysis into an executive summary Once the investigation is complete, the same agent can be prompted to produce a short summary for leadership. This is an important part of the workflow because the technical investigation and the business explanation are different tasks. The underlying work may involve SQL, schemas, customer records and campaign files. The final audience may simply need to know: **What changed? Why did it change? And does it represent a genuine business problem?** The example shows the agent generating an executive summary after completing the investigation. ## From one-off investigation to a reproducible data pipeline Google Cloud's example does not end with root cause analysis. The next step is to turn the investigation into something reusable. The user asks the agent to build a persistent **dbt project** that joins BigQuery staging models with Cloud SQL customer and pet profile attributes. The prompt also asks for a uniqueness test on `order_id` and a `dbt build`. The agent creates a virtual Python environment with `dbt-bigquery`, writes the project models and tests, and runs the build. But the first attempt fails. The uniqueness test identifies duplicate `order_id` values. The underlying issue is that customers can have multiple pets. The initial model attached pet profiles directly to orders, meaning an order belonging to a household with three pets could become three rows. The agent reads the terminal output, identifies the problem, modifies the dbt logic and reruns the build until the test passes. This part of the example highlights an important limitation of agentic data workflows. ## Agents can write code, but data quality still matters Generating code is not the same as producing a trustworthy data pipeline. The Data Agent Kit example deliberately includes a failed build and a data-quality test. That is significant because automated data work still needs validation. Tests can identify problems that may not be obvious from the generated SQL or the final output. In this case, the uniqueness test exposed a modeling issue that could have affected downstream results. Google's example therefore presents data quality checks as an important part of the agentic workflow rather than something that disappears because an agent is doing the implementation. ## ![info-3](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1788936875164-info-3--110-.webp) ## What this means for data teams The broader significance of the Data Agent Kit is less about eliminating SQL and more about changing how analysts and developers interact with data systems. Traditional analysis often involves a sequence like this: ``` Question → identify system → inspect schema → write query → execute → inspect result → switch system → write another query → combine findings → explain result ``` An agentic workflow can compress much of that process into a conversational investigation. The developer still defines the question, reviews the work and determines whether the conclusion makes sense. But the agent can handle some of the repetitive interaction with schemas, queries, tools and development environments. The announcement highlights a broader industry shift toward **agentic data workflows**, where AI systems are increasingly being positioned as interfaces for interacting with enterprise data and development tools. For enterprises, this could mean that the value of data agents will depend not only on their ability to generate code or SQL, but also on how effectively they can work with real data environments, follow defined permissions and operate alongside testing and governance practices. These are broader implications rather than claims made by Google about specific enterprise outcomes. ## Why MCP and skills matter The Data Agent Kit also illustrates two architectural ideas that are becoming increasingly relevant in agentic development. **MCP provides connectivity.** It gives agents a standardized way to interact with external tools and data sources. **Skills provide context.** They give the agent additional instructions about how a particular environment or workflow should be handled. That separation can make an agentic workflow more modular. Instead of building every capability into one large application, developers can connect tools through MCP and provide additional workflow knowledge through editable skills. Google describes this combination as a way to reduce the friction between a business question and the systems required to answer it. ## Data Agent Kit is currently in preview Google Cloud says the Data Agent Kit is currently **in preview**. According to the announcement, it works with Antigravity 2.0, Antigravity CLI, Antigravity IDE, Claude Code, Codex, Cursor and other popular tools. Google also provides three routes for exploring the technology: - A codelab covering the [Analytics with Data Agent Kit and Antigravity IDE scenario](https://codelabs.developers.google.com/dak-analytics-eng-antigravity-ide#0) - Google Cloud documentation for the Data Agent extension - An open-source [GitHub](https://github.com/gemini-cli-extensions/data-agent-kit-starter-pack)repository containing the skills and tools Additional details were not disclosed in the announcement regarding broader production availability or specific enterprise deployment requirements. ## The bigger picture The most interesting part of Google's Data Agent Kit example is not that an AI agent can write SQL. Modern AI coding tools already demonstrate that capability. The more important shift is the ability to connect **investigation, execution and development** within the same workflow. A question that begins as "Why did this metric change?" can lead to analytical queries, operational database lookups, campaign-file inspection, executive communication and eventually a reproducible data model. That creates a different model for data work. The analyst remains responsible for directing the investigation and validating the result. The agent takes on more of the mechanical work involved in moving between systems and implementing the analysis. The example also makes clear that agentic analytics does not remove the need for data quality checks. In fact, as agents take on more implementation work, automated tests and human review become important parts of making those workflows dependable. For data teams experimenting with agentic development, that may be the most practical takeaway: **the goal is not simply to ask AI for an answer. It is to give an agent the tools, context and validation mechanisms required to investigate how the answer was reached.** Google Cloud's Data Agent Kit is an example of that approach, bringing those capabilities directly into the developer environment. ## Original source https://cloud.google.com/blog/products/data-analytics/agentic-analytics-with-the-data-agent-kit ## Tags `#DataAnalytics` · `#AgenticAI` · `#GoogleCloud` · `#DataAgents` · `#BigQuery` · `#ModelContextProtocol` · `#MCP` · `#AIEngineering` --- ## About this content This Markdown news article is the citation-grade twin of [Google Cloud's Data Agent Kit Brings Agentic Analytics Into the IDE](https://www.xcademia.com/news/google-cloud-s-data-agent-kit-brings-agentic-analytics-into-the-ide). It is published by **Xcademia** (UK Companies House 12322710) and is available for AI search engines and large language models to index, summarise, and cite. When citing or quoting, please attribute *Xcademia* and link back to the source URL above. - Source: https://www.xcademia.com/news/google-cloud-s-data-agent-kit-brings-agentic-analytics-into-the-ide - Publisher: Xcademia — https://www.xcademia.com - Catalogue index: https://www.xcademia.com/llms-full.txt