Google Cloud releases Data Agent Kit for agentic analytics in IDEs
Google Cloud introduced the Data Agent Kit, a collection of MCP servers and markdown skills that connect IDE agents to BigQuery, Cloud SQL, and Cloud Storage. Available as a VS Code-fork extension and as a plugin for Claude Code, Codex, and Antigravity tools, it lets agents execute SQL after permission prompts and expose an audit trail of tool calls.
Google Cloud has released the Data Agent Kit, a set of MCP servers and agent skills that helps data developers run data workflows from their IDEs. The company frames the kit as a response to open-ended questions that have no single dashboard, such as a 7 percent drop in average order value in January while total revenue stayed flat. Pieces of those answers often sit in a data warehouse, a production PostgreSQL instance, and raw JSON in an object store, forcing practitioners to write similar queries across dialects and bounce between browser tabs. The kit is available as an extension for VS Code forks including Antigravity IDE and Cursor, and as a plugin for Antigravity 2.0, Antigravity CLI, Claude Code, and Codex. The kit relies on two core mechanisms. Model Context Protocol is an open standard that connects an agent to tools, databases, and remote cloud infrastructure. Skills are markdown files that augment the agent's knowledge and teach it how to interact with a specific stack. Instead of generating SQL snippets and copy-pasting them into a console, Data Agent Kit lets agents run the queries and read the results on the user's behalf. In the sample architecture described by Google Cloud, sales history sits in BigQuery, live customer records live in a Cloud SQL PostgreSQL instance, and marketing campaign rules are raw JSON files in Cloud Storage. The announcement walks through that average-order-value scenario in the IDE chat pane, starting with a natural-language prompt to confirm baseline numbers. The agent invokes relevant skills and prepares to query data, but the IDE pauses to ask for permissions to use MCP tools such as execute_sql_readonly. The user can allow a tool once for auditing or select always allow. Agentic IDEs let developers inspect the execution trail, including each MCP tool call and the raw SQL sent to BigQuery. Google Cloud notes it is important to keep an eye on generated code, though reading a query can take much less time than writing one.