1|# [Community Project] `create-google-adk-agent` — one-command scaffold for ADK projects
2|
3|## 👋 Hey ADK community!
4|
5|I wanted to share a project I've been building that I think fills a real gap in the ADK developer experience: **`create-google-adk-agent`** — a CLI scaffold tool (think `create-next-app`, but for Google ADK agents).
6|
7|- 📦 npm: https://www.npmjs.com/package/create-google-adk-agent
8|- 🐍 PyPI: https://pypi.org/project/create-google-adk-agent/
9|- 💻 GitHub: https://github.com/unrealandychan/create-adk-agent
10|
11|---
12|
13|## 🤔 The Problem
14|
15|Getting started with a new ADK project today involves a fair bit of ceremony: manually creating the directory structure, writing boilerplate `__init__.py` and `agent.py` files, wiring up environment variables, choosing and configuring an LLM provider, and — if you want to go beyond Google AI Studio — navigating LiteLLM or Vertex AI setup from scratch.
16|
17|There's no official `adk init` or scaffold CLI for local development, especially for developers who aren't yet on GCP or want to prototype quickly with OpenAI, Anthropic, or a local Ollama model. This friction is real for newcomers and even experienced practitioners spinning up new projects.
18|
19|---
20|
21|## ✨ What `create-google-adk-agent` Does
22|
23|It's a single command that walks you through an interactive setup and generates a fully runnable ADK project:
24|
25|```bash
26|# via npm (no install needed)
27|npx create-google-adk-agent
28|
29|# or via Python
30|uvx create-google-adk-agent
31|# or: pip install create-google-adk-agent && create-google-adk-agent
32|```
33|
34|### Interactive flow
35|
36|```
37|✔ Project name: my-research-agent
38|✔ Agent type: multi-agent
39|✔ LLM provider: OpenAI (via LiteLLM)
40|✔ Model: gpt-4o
41|
42|✅ Project created! Run:
43| cd my-research-agent
44| uv run adk web
45|```
46|
47|### Supported agent types
48|| Type | Description |
49||---|---|
50|| `single` | A single root agent |
51|| `multi-agent` | Orchestrator + sub-agent scaffold |
52|| `agentic_rag` | RAG pipeline with document ingestion wired in |
53|
54|### Supported LLM providers
55|- **Google AI Studio** (Gemini)
56|- **Vertex AI**
57|- **OpenAI** (via LiteLLM)
58|- **Anthropic** (via LiteLLM)
59|- **Ollama** (local models)
60|
61|### Built-in tools (opt-in during setup)
62|- 🌐 **Web Fetch** — fetch and parse web pages
63|- 📄 **File Reader** — read PDF and `.docx` documents
64|- 🐍 **Python Code Executor** — sandboxed code execution
65|- 🕐 **Datetime** — current date/time utilities
66|
67|### MCP server support
68|Auto-generates `mcp.json` for: `filesystem`, `fetch`, `github`, `memory`, `slack`
69|
70|### Other niceties
71|- `--yes` / `-y` flag for fully non-interactive / CI mode
72|- Copilot instructions pre-wired (`.github/copilot-instructions.md`)
73|- `uv` + `adk web` ready out of the box
74|
75|---
76|
77|## 💡 Two Things I'd Love to Discuss
78|
79|### 1. Add to a community projects section
80|If this repo has (or plans to have) a community projects listing in the README or a dedicated `COMMUNITY.md`, I'd love for `create-google-adk-agent` to be included. I think it genuinely lowers the barrier to entry for new ADK developers.
81|
82|### 2. Contribute the built-in tools to `tools/`
83|The four built-in tools (web fetch, file reader, Python executor, datetime) were written to be drop-in ADK-compatible tools. I'd be happy to contribute them as proper community tools under a `tools/` module in this repo — with tests, type annotations, and documentation — if that's something the maintainers would find valuable. They cover some of the most common "first things I reach for" when building agents.
84|
85|---
86|
87|## 🙏 Thanks
88|
89|Thanks for building ADK and fostering this community space. Happy to answer any questions, take feedback, or open a separate PR for the tools contribution if there's interest!
90|
91|— Andy ([@unrealandychan](https://github.com/unrealandychan))
92|
