Google Data Agent Kit: Point Your Agent at Data (2026)

If you work with data on Google Cloud, you know the ritual. Your coding agent writes beautiful SQL — against a schema it has never seen. So you paste in CREATE TABLE statements, sample rows and yesterday’s failed-job log. The agent reads them, nods along, and promptly forgets half of it. Rinse, repeat, every session.

Google Data Agent Kit ends that ritual. Announced as generally available in the week of October 5–9, 2026 on the Google Cloud Blog, the Google Data Agent Kit is a free set of Model Context Protocol (MCP) tools and Google-authored agent skills that lets the coding agent you already use — Claude Code, Codex, Cursor, VS Code, Antigravity — work directly against your live Google Cloud data products.

This guide covers what the Google Data Agent Kit actually is, what’s new in the GA release, how to install it in under a minute, what you can build with it, and the honest caveats you should know before pointing an agent at production data. Verified October 14, 2026.

Table of Contents

The problem Google Data Agent Kit solves

Coding agents are remarkably good at writing SQL, PySpark and pipeline code. What they don’t have by default is context about your environment: which tables exist, how they’re partitioned, which ones your team trusts, and why last night’s job failed. Without that context, even a strong agent works from assumptions, and you end up pasting schemas and error logs into chat to fill in the gaps.

The Data Agent Kit fills the gap natively. Your agent can inspect schemas, run queries, read job logs and manage resources in the live environment — with every step running under your own IAM permissions. The pitch, in one sentence: stop pasting schemas into chat; point your agent at the warehouse.

What the kit is: MCP tools plus Google-authored skills

The Google Data Agent Kit has two halves, and you need both:

MCP tools. Connections to more than 15 Google Data Cloud services, so your agent can inspect schemas, run queries, read job logs and manage resources in your live environment. BigQuery, BigQuery Graph, Spanner, AlloyDB, Cloud SQL, Bigtable, Managed Service for Apache Spark, Dataform, Managed Service for Apache Airflow, the open Lakehouse and the Knowledge Catalog are all covered.

Google-authored skills. Open-source instructions written by Google Cloud engineers that teach your agent data best practices — like optimising BigQuery SQL, designing Bigtable row keys and building dbt (data build tool) pipelines. Because the skills are open source, you can audit them, fork them, and write custom ones for your own internal standards. If you’ve built your own MCP server before — we wrote a complete guide to building one in Laravel — the client side of the equation will feel familiar.

Google’s own walkthrough example shows the intended workflow. Ask your agent: “Forecast next month’s demand for our top-selling products and check whether we have enough inventory to meet it.” The kit loads the relevant skills, searches the Knowledge Catalog for the sales and inventory tables your team trusts, runs the forecast in BigQuery, checks current stock levels in AlloyDB for PostgreSQL, and brings the combined answer back to your editor or terminal. Every step runs with your own IAM permissions.

What’s new in the GA release

Three headline additions shipped with GA, plus dozens of quality-of-life improvements:

Component	What GA added
BigQuery Graph	Property-graph support: the `bigquery-graph-author` skill maps your tables to nodes and edges, validates each proposed relationship against the actual data, and shows you a plan to approve before creating anything. It can even start from an ER diagram you already have. The `bigquery-graph-query` skill then writes the GQL, and the graph visualizer in the IDE lets you click through the results.
Bigtable	Fully supported database, alongside Spanner, AlloyDB and Cloud SQL. The `bigtable-basics` skill designs your schema around how the data will be read and flags hotspots and full table scans before you create anything. Your agent can then create the table and query it with GoogleSQL, with column families flattened into readable columns.
Managed Service for Apache Spark on the open Lakehouse	Your agent could already query the Apache Iceberg tables in your Lakehouse through BigQuery. Now it can work with those same tables using serverless Spark, with no cluster to manage. Each session keeps its state, so temporary views carry across statements, and you get Iceberg’s full feature set: branching, time travel and schema evolution.

The federate-lakehouse-catalog skill connects your Lakehouse to AWS Glue and Databricks Unity Catalog, so your agent queries that data in place — no ingestion pipeline to build first.

On the experience side, GA streamlines setup and daily work. You sign in once and select the Google Cloud services you use; the kit automatically enables the required APIs, installs the matching skills and configures your MCP servers with no manual setup files. Inside the IDE, the SQL editor and notebooks support inline code generation, @ references to tables and diff views for suggested changes. The extension shares your active project, open file and the error from the query you just ran with your agent, so asking it to “fix this query” just works. New Spark notebooks automatically create and select a Spark Connect runtime, Spark SQL queries in the editor run in isolated sessions with built-in execution metrics, the catalog explorer loads faster and includes BigQuery public datasets in the sidebar, and tuned notebook skills finish notebook tasks using fewer tokens.

How to install Google Data Agent Kit

Google says setup takes under a minute, and the path you take depends on where you already work:

Option 1: the IDE extension

Search for “Google Cloud Data Agent Kit” in the Extensions panel of VS Code, Antigravity IDE, Cursor or any VS Code-compatible editor. It is also installable from the VS Code Marketplace or Open VSX. In Antigravity 2.0, go to Settings > Customizations > Build with Google Plugins and download it. In Cloud Shell and Cloud Workstations it comes pre-installed — just open the editor and sign in.

The extension also brings a lightweight version of the Google Cloud console into your editor, so you can browse data, run queries and review your agent’s work without switching windows.

Option 2: the CLI plugin

If your workflow lives in the terminal — Claude Code, Codex or the Antigravity CLI — install the plugin from the official GoogleCloudPlatform/data-agent-kit-plugin GitHub repository. The commands for each agent are listed on the repo. It slots into the same workflow if you already run open models inside Claude Code with Together Link.

Four things worth building first

Schema discovery without the copy-paste. Ask it to list your BigQuery datasets, explain which tables are partitioned and how, and which ones the team’s docs treat as source of truth. This is the fastest way to feel what “context” is actually worth.
A demand forecast, end to end. The official example: Knowledge Catalog for trusted tables, a BigQuery forecast, an AlloyDB stock check, the answer in your editor. Watch how many steps it completes before asking you for input.
A BigQuery property graph. Describe customers, orders and suppliers as a graph, review and approve the plan, then query it with GQL the agent wrote. You never hand-write a CREATE PROPERTY GRAPH statement or learn GQL just to explore relationships.
Pipeline diagnosis. Once your logic works, the agent turns it into dbt or Dataform models and the gcp-pipeline-orchestration skill schedules them with your notebooks as an Orchestration Pipeline on Managed Service for Apache Airflow. When a run fails, click Diagnose on the visual pipeline canvas in the IDE — the Airflow and Spark troubleshooting skills trace the root cause and propose a fix for you to approve.

The honest caveats

Free kit, paid queries

The kit is included at no additional cost — you pay standard pricing only for the Google Cloud services your agent uses. The skills steer the agent toward cost-aware query patterns, like checking partition keys and running a dry run before executing a BigQuery query to avoid accidental full-table scans. For practice, the BigQuery sandbox lets you experiment without billing; on-demand queries start at $6.25 per TiB scanned with the first 1 TiB each month free (confirm current pricing on the official BigQuery pricing page before budgeting).

A dry run before any agent-written query is a habit worth enforcing:

bq query --use_legacy_sql=false --dry_run "SELECT order_id, SUM(total) FROM `my-project.sales.orders` WHERE order_date >= DATE_SUB(CURRENT_DATE(), INTERVAL 30 DAY) GROUP BY 1"

Skills are guidance, not a security boundary

The agent connects as you or as a service account you impersonate, so row- and column-level security policies apply automatically — that’s the good part. But agent skills can only instruct the model to follow your standards; they cannot enforce access restrictions. IAM is what actually gates who can touch what.

Practical adoption order: pick a development project, give the agent the smallest permission set that supports the job, start with read-only tasks like schema discovery and query review, and validate suggested changes in a non-production environment before expanding permissions. Review the diff views before anything mutates production. Admins can govern MCP access with IAM, screen MCP traffic with Model Armor, and scope agents with VPC Service Controls and Principal Access Boundary policies.

Auto-setup touches your project

Signing in once and letting the kit enable APIs and install skills is genuinely convenient — just make sure the selected Google Cloud project is the one you intend. Keep experiments off the production billing project until you’ve seen what the kit enables and what the agent does with it.

Three first prompts to try today

“List my BigQuery datasets and tell me which tables are partitioned, and by what.”

“Dry-run this query and estimate its cost before running it.”

“Look at the last failed run of my Airflow DAG and propose a fix — don’t change anything until I approve.”

Google Data Agent Kit Is GA: Point Your Coding Agent at the Live Warehouse (2026)