ADK Tools

Build, run, and deploy Google ADK agents — without touching the terminal.

A VS Code, Cursor, and Windsurf extension that brings the full Google Agent Development Kit workflow into your editor: one-click servers, model switching, deployment to Cloud Run and Vertex AI, eval runner, auth setup, and code snippets.

Preview

ADK Tools — server running with syntax-highlighted agent.py and live status bar

Web UI running on :8000 · live status bar · syntax-highlighted agent.py

Sidebar with project info, running state, and sections · Model switcher with current model highlighted

Auto-detects uv / pip / pipenv and installs google-adk in one click

What it does


⚡ One-click servers	Start `adk web` or `adk api_server` with port conflict detection, hot reload, and status bar live state
🔄 Model switcher	Detect and rewrite `model=` in your agent file — Gemini 2.0/2.5, Claude via Vertex AI
☁️ Deploy anywhere	Cloud Run, Agent Engine, and GKE via native `adk deploy` — with guided step-by-step inputs
🧪 Eval runner	Discover eval files and run `adk eval` or generate cases with `adk eval_set`
🔑 Auth wizard	Set up Gemini API key (writes to `.env`, checks `.gitignore`) or Vertex AI credentials
📦 Project scaffolding	3-step wizard using `adk create` — name, model, output folder
💻 CLI runner	`adk run` with multi-agent picker and session mode selection
📝 Code snippets	Python & TypeScript snippets for agents, tools, callbacks, pipelines

Installation

Open VSX (Cursor, Windsurf, Gitpod) Search ADK Tools in the Extensions panel, or:

ext install lenixbyte.adk-tools

Manual (.vsix) Download from Releases and drag into the Extensions panel.

Requirements

VS Code 1.85+ / Cursor / Windsurf
ADK CLI — pip install google-adk

Features

Open the ADK icon in the Activity Bar to see your project at a glance.

ADK Project
├── my_agent          python · uv          ← project row
│   └── gemini-2.0-flash  click to switch  ← model row
├── Development  (expanded)
│   ├── ▶ Run Web UI
│   ├── ⊙ Run API Server
│   ├── ⬛ Run CLI (adk run)
│   ├── 🔑 Auth Setup
│   └── ⚙ Run Options
├── Deployment  (collapsed)
│   ├── ☁ Deploy to Cloud Run
│   ├── ⬡ Deploy to Agent Engine
│   └── ⎔ Deploy to GKE
├── Evaluation  (collapsed)
│   ├── ▷ Run Eval
│   └── + Generate Eval Cases
└── Help & Diagnostics  (collapsed)

Status Bar

Three persistent items appear when an ADK project is detected:

State	Status Bar
Idle	`⚡ ADK · uv` `▶ Web` `⊙ API`
Web running	`⚡ ADK · uv` `📡 :8000` (→ browser) `◼ Stop`
API running	`⚡ ADK · uv` `⊙ :8000` (→ browser) `◼ Stop`
Hot reload	`↻ hot reload` (yellow badge)

The extension polls the port every 4 seconds — if your server exits via Ctrl+C, the UI resets automatically.

Server Management

ADK: Run Web UI / ADK: Run API Server

Detects your environment (pipenv / uv / .venv / system) and builds the right command
Port conflict modal: Kill & Restart, Open Browser, or Show What's Running
Web and API share port 8000 — starting one gracefully stops the other

Run Options (persisted per workspace)

Setting	Flag	Default
Port	`--port`	`8000`
Hot Reload	`--reload_agents`	off
Log Level	`--log_level`	`INFO`
Session Storage	`--session_service_uri`	`memory://`

Session backends: memory:// · sqlite:///./sessions.db · agentengine://

Model Switcher

ADK: Switch Model reads model= from your agent.py, marks the current model with ← current, and rewrites the file in-place. Offers to restart the server after switching.

Supported models

Model	Notes
`gemini-2.5-pro`	Most capable
`gemini-2.0-flash`	Recommended default
`gemini-2.0-flash-lite`	Fastest / lowest cost
`gemini-1.5-pro` / `gemini-1.5-flash`	Previous generation
`claude-3-5-sonnet@20241022`	Anthropic via Vertex AI
`claude-3-haiku@20240307`	Anthropic via Vertex AI (fast)

Auth Setup

Gemini API Key — enters key, writes GOOGLE_API_KEY to .env, verifies .gitignore covers it. Opens AI Studio if you need a key.

Vertex AI — runs gcloud auth application-default login, sets GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_LOCATION in .env.

Deployment

ADK: Deploy Agent → pick target:

Cloud Run (4 steps) — project, region, service name, with/without Web UI

adk deploy cloud_run --project P --region R --service_name S [--with_ui] .

Agent Engine (3 steps) — project, region, display name

adk deploy agent_engine --project P --region R --display_name "N" --validate-agent-import .

GKE (4 steps) — project, region, cluster name, service type

adk deploy gke --project P --region R --cluster_name C --service_type T .

Code Snippets

Type a prefix in .py or .ts files and press Tab.

Python

Prefix	Inserts
`adk-agent`	`LlmAgent(...)`
`adk-root`	`root_agent = LlmAgent(...)`
`adk-tool`	Tool function with full docstring
`adk-before-model`	`before_model_callback`
`adk-after-model`	`after_model_callback`
`adk-before-tool`	`before_tool_callback`
`adk-sequential`	`SequentialAgent(...)`
`adk-parallel`	`ParallelAgent(...)`
`adk-loop`	`LoopAgent(...)`
`adk-state`	Session state read/write
`adk-agent-file`	Complete `agent.py` template

TypeScript

Prefix	Inserts
`adk-agent`	`new LlmAgent({...})`
`adk-root`	`export const rootAgent = ...`
`adk-tool`	`new FunctionTool({...})`
`adk-sequential`	`new SequentialAgent({...})`
`adk-parallel`	`new ParallelAgent({...})`

Environment Detection

The extension detects your Python environment automatically — no configuration required.

Environment	Detected by	Command
Pipenv	`Pipfile` present	`pipenv run adk ...`
uv	`uv.lock` or `[tool.uv]` in pyproject	`uv run adk ...`
.venv	`.venv/bin/adk` exists	`/path/.venv/bin/adk ...`
System	fallback	`adk ...`

Project Detection

The extension recognizes ADK projects by checking (in order):

.adk/ directory at workspace root
agent.py with ADK imports (google.adk, LlmAgent, etc.) within 3 directory levels
requirements.txt or pyproject.toml containing google-adk

Multi-agent repos are supported — the CLI runner lets you pick which agent to target.

All Commands

Access from the sidebar, status bar (⚡ ADK → command menu), or Ctrl+Shift+P → type ADK.

Command	Description
`ADK: Run Web UI`	Start `adk web`
`ADK: Run API Server`	Start `adk api_server`
`ADK: Stop All Servers`	Stop running servers
`ADK: Open ADK Web UI in Browser`	Open localhost
`ADK: Run CLI (adk run)`	Interactive CLI with mode picker
`ADK: Switch Model`	Change model in `agent.py`
`ADK: Run Options`	Configure port / hot reload / log level
`ADK: Auth Setup`	Gemini API key or Vertex AI setup
`ADK: Deploy Agent`	Cloud Run / Agent Engine / GKE
`ADK: Run Eval`	Run `adk eval`
`ADK: Generate Eval Cases`	Run `adk eval_set generate_eval_cases`
`ADK: Create New Agent Project`	Scaffold with `adk create`
`ADK: Open Agent File`	Open `agent.py`
`ADK: Edit .env File`	Open or create `.env`
`ADK: Show .env Summary`	Preview with masked secrets
`ADK: Run Diagnostics`	Check tools, port, environment
`ADK: Kill Port 8000`	Force-free port 8000
`ADK: Show Output Log`	Open ADK Tools output channel
`ADK: Getting Started`	Getting started panel
`ADK: Open ADK Documentation`	Open adk.dev

License

Apache 2.0 — see LICENSE

Made with ♥ for the Google ADK community