P3.3: Research & Prototype Advanced ADK Concepts

[Virgo-Alpha]

This is a research-focused or "Spike" ticket aimed at improving our agent architecture for future phases. The goal is to explore advanced features of the Agent Development Kit (ADK) to understand how they can make our multi-agent system more powerful and efficient. This work will be done in small, isolated proof-of-concept scripts, not integrated into the main application yet.

Research Areas & Prototypes

Model-Context Protocol (MCP) for Tool Access (F2.8.1)

• Goal: The ADK provides a feature called MCP that allows agents to access tools and data in a more structured way. The goal is to understand how this works.
• Task: Create a small, standalone Python script that defines a simple tool. Instead of passing the tool directly to the agent, investigate how to register it with the MCP and have the agent use it from there. This could be useful for managing a large, shared library of tools in the future.

Direct Agent-to-Agent (A2A) Communication (F2.8.2)

• Goal: Our current architecture uses API calls for inter-agent communication. The PRD mentions A2A communication. We need to understand what this is and how it differs.
• Task: Investigate the sub_agents parameter for the Agent class (the "in-process" orchestration we tested). Create a small script that defines two agents and makes them communicate using this sub_agents feature. Document the pros and cons compared to our current microservices approach. This will inform our decision on whether to use this pattern for more tightly-coupled agent interactions in the future.

ADK Web UI for Observability (F2.8.3)

• Goal: To become more efficient at debugging our agents.
• Task: Thoroughly explore the local ADK Playground UI (make playground). Experiment with the step-by-step execution trace, inspect the LLM requests and responses, and document best practices for using this tool to quickly identify issues in an agent's reasoning loop.

Acceptance Criteria (Checklist)

• A brief document or README is created summarizing the findings for MCP.
• A standalone script demonstrating A2A communication using sub_agents is created.
• A summary of the pros and cons of A2A vs. our current microservice architecture is written.
• The team feels confident in using the ADK Playground for future agent development and debugging.