GCP Multi-Agent Demo (using Google ADK)

This is a small project that demonstrates a multi-agent system built with the Google Agent Development Kit (ADK). It uses a MetaAgent to coordinate tasks between specialized agents that interact with various tools.

Overview

The system features:

• MetaAgent: Uses Gemini and routes user requests to the appropriate specialized agent or uses tools directly.
• ResourceAgent: Manages Google Compute Engine VM instances (create, delete, list, start, stop, get details) through the GCE API.
• DataScienceAgent: Interacts with Google BigQuery (runs queries, creates datasets).
• GithubAgent: Interacts with GitHub via it's official MCP server.
• MistralChatAgent: A Mistral-Small agent integrated via LiteLLM for general conversation.
• Currency Conversion: The MetaAgent uses a tool to communicate via A2A with an external LangGraph-based agent (running as a separate service) to get currency exchange rates using the Frankfurter API and Gemini.
• Streamlit UI: Provides a web interface for interacting with the agents.
• Deployment Scripts: Automates building and deployment using Google Cloud Build to deploy on Google Cloud Run using a Cloud Build Trigger on the GitHub repo. Container images are stored in Google Artifact Registry.

Project Diagram : Workflow of the different component

Project Structure

adk-demo/
├── agents/                     # Agent definitions and tools
│   ├── datascience/
│   │   ├── agent.py            # DataScience Agent
│   │   └── tools.py            # BigQuery tools
│   ├── githubagent/
│   │   ├── agent.py            # GitHub Agent
│   │   └── tools.py            # GitHub MCP Tools
│   ├── langgraphagent/
│   │   └── tools.py            # LangGraph A2A Tool
│   ├── meta/
│   │   └── agent.py            # Meta Agent (orchestrator)
│   └── resource/
│       ├── agent.py            # Resource Agent
│       └── tools.py            # Compute Engine tools
├── assets/
│   └── td-flow-chart.png       # Diagram
    └── td-flow-chart.mermaid   # Mermaid TD flow chart
├── ui/                         
│   └── app.py                  # Streamlit UI application
├── requirements.txt            # Python dependencies for main app
├── Dockerfile                  # Dockerfile for main ADK/Streamlit app
├── Dockerfile.langchain        # Dockerfile for external LangGraph agent
├── Dockerfile.mcp              # Dockerfile for MCP Server (if used)
├── cloudbuild.yaml             # Cloud Build configuration
├── setup_trigger.sh            # Script to setup Cloud Build trigger
├── .env.example                # Example environment file
├── .env                        # Local environment file (gitignored)
├── .gitignore                  # Files ignored by Git
├── LICENSE                     # Project license
└── README.md                   # This file

Prerequisites

• Bash
• A Google Cloud Project with the following APIs enabled:
  • Compute Engine API
  • BigQuery API
  • Cloud Build API
  • Cloud Run API
  • Artifact Registry API
  • AI Platform (Vertex AI) API (for Gemini / Mistral models)
• A user with Owner role on the project (or sufficient permissions for the APIs above)
• An Artifact Registry Docker repository
• A Google API Key (for ADK interaction with Gemini models).

Setup

1. Clone the repository:

   git clone <your-repository-url>
   cd <your-repo-directory>
   

2. Set Environment Variables:

   • Copy the .env.example file in the project root to a new .env file and set your specific variables (Project ID, Region, API Keys, etc.). Ensure GOOGLE_API_KEY and GCP_PROJECT_ID are set.
   • Security Note: For production, manage sensitive keys (GOOGLE_API_KEY, GITHUB_TOKEN) securely using tools like Secret Manager.

3. Make script executable:

   chmod +x setup_trigger.sh && ./setup_trigger.sh
   

This script creates/updates a Cloud Build Trigger. When triggered (e.g., by a push to the main branch), Cloud Build will build container images using the Dockerfiles and deploy the services (Streamlit/ADK App, MCP Server, LangGraph Agent) to Cloud Run based on cloudbuild.yaml. The script will also create Substitutions in your Cloud Build Trigger setting all you .env variables.

Usage

A single Streamlit interface is accessible via your Cloud Run service name. Navigate to the Google Cloud Console, Cloud Run, your-streamlit-service and click on the link at the top. Your URI should look like https://adk-multi-agent-streamlit-1234567890.us-central1.run.app

Configuration Notes

• Session Persistence: The Streamlit UI uses InMemorySessionService by default. Chat history will be lost if the Cloud Run instance restarts. For persistent sessions, explore other ADK session services documented in the Agent Development Kit.

Contributing

Contributions and feedback are welcome!