内容
# Build a Multi-Agent Creative Studio with Google's Agent Stack: ADK, A2A, MCP on Cloud Run & Agent Runtime
A hands-on codelab for building a distributed multimodal multi-agent system using **Google ADK**, **A2A protocol**, **MCP**, and **Gemini Enterprise Agent Platform Runtime**. Participants build a complete Instagram campaign generator from scratch, deploying five specialist agents that collaborate through A2A communication.
## What You Build
A distributed multi-agent creative studio where specialized AI agents collaborate to produce complete Instagram campaigns:
| Agent | Role |
|---|---|
| **Brand Strategist** | Market research, competitor analysis, audience insights |
| **Copywriter** | Instagram captions using ADK Skills (platform guidelines + caption formulas) |
| **Designer** | Visual concepts + real image generation via Gemini, stored in GCS |
| **Critic** | Quality review with structured APPROVED / NEEDS_REVISION scores |
| **Project Manager** | Campaign timeline, tasks, and optional Notion integration via MCP |
Coordinated by a **Creative Director** orchestrator that sequences the agents, handles the Critic's revision loop, and compiles the final campaign.
## Key Concepts Covered
- Building ADK agents with tools, callbacks, and system instructions
- **ADK Skills** - packaging reusable knowledge into modular files loaded on demand
- **Multimodal** - bridging a text agent to an image model via a `FunctionTool`
- **A2A protocol** - agents communicating over HTTPS as independent services
- **MCP toolsets** - connecting agents to external services (Notion) without custom glue code
- **`after_tool_callback`** - intercepting tool responses for error handling and schema injection
- Deploying agents to **Cloud Run** and **Gemini Enterprise Agent Platform Runtime**
## Repository Structure
```
workshop/
starter/ ← participant starting point (TODOs to fill in)
agents/
brand_strategist/
copywriter/ ← includes ADK Skills (instagram-copywriting)
designer/ ← Gemini image generation + GCS upload
critic/
project_manager/ ← Notion MCP integration + error handling callback
creative_director/ ← orchestrator with Critic revision loop
deploy/ ← Cloud Run + Gemini Enterprise Agent Platform deployment scripts
docs/ ← published codelab (codelab.json + index.html)
diagrams/ ← screenshots and GIFs used in the codelab
```
## Prerequisites
- Google Cloud project with billing enabled
- APIs: Vertex AI, Cloud Run, Cloud Storage, Secret Manager
- `gcloud` CLI authenticated
## Getting Started
```bash
git clone https://github.com/Saoussen-CH/mas-a2a-gcp.git
cd mas-a2a-gcp/workshop/starter
uv sync
cp .env.example .env
# Fill in .env with your project values
uv run adk web agents --allow_origins='*'
```
Full step-by-step instructions are in the published codelab at [codelabs.developers.google.com/ai-creative-studio-adk-a2a](https://codelabs.developers.google.com/ai-creative-studio-adk-a2a).
## Tech Stack
- [Google ADK](https://adk.dev) `1.31.1`
- [A2A Protocol](https://github.com/google/A2A)
- Gemini models (text + image generation) on Vertex AI
- Cloud Run, Gemini Enterprise Agent Platform Runtime, Cloud Storage
- MCP (Model Context Protocol) for Notion integration
MCP 市场