Project Horizon

Exploring the Future of Interoperable, Multimodal AI Agent Systems

---

Overview

Project Horizon explores the future of collaborative, multimodal AI agent systems. It investigates how sophisticated, responsive, and interoperable agents can be built by integrating cutting-edge agent development frameworks and open communication protocols.

(This diagram illustrates the general interaction patterns explored in Project Horizon, involving a Host Agent, Specialist Agents, and Tool Servers communicating via A2A and MCP.)

This project serves as a testbed and a collection of reference implementations for:

• Integrating diverse agent frameworks and tools like Google's Agent Development Kit (ADK) and the Gemini Live API.
• Implementing standardized communication patterns between independent agents using protocols like Agent2Agent (A2A).
• Enabling agents to securely interact with external tools and data sources using protocols like the Model Context Protocol (MCP).
• Exploring advanced agent architectures, design patterns, and evaluation techniques.

Project Horizon aims to capture and demonstrate the future of agent interactions, providing technical implementations, architectural blueprints, and practical design patterns for building the next generation of AI applications.

Core Technologies

This project showcases and explores the integration of several key technologies:

• Google Agent Development Kit (ADK): For building user-facing agents, orchestrating tasks, and managing multimodal communication.
• Gemini Live API: Powers real-time, low-latency, bidirectional voice (and potentially video) interactions within ADK.
• Agent2Agent Protocol (A2A): An open standard for communication and task delegation between independent AI agents.
  • This repository includes examples using a custom A2A implementation (for historical reference). An implementation with the official google-a2a-python SDK is planned.
• Model Context Protocol (MCP): An open standard allowing agents to securely interact with dedicated "Tool Servers" that provide specific functionalities or data.

Example Applications

This repository contains a growing collection of example applications, each demonstrating different aspects of multi-agent systems. Each example is self-contained within the examples/ directory and includes its own specific setup instructions.

Available Examples:

1. Synchronous Stock Lookup (Custom A2A/MCP Implementation)
   • Directory: examples/stock_lookup_custom_a2a/
   • Description: This is the original proof-of-concept demonstrating a real-time stock price lookup using ADK Live, a custom A2A implementation for delegation, and MCP over stdio for tool execution. It is preserved with its original dependencies for reproducibility.
   • Technologies: ADK Live, Custom A2A, Custom MCP Client, Finnhup API.

More examples demonstrating various agent architectures and protocol features will be added over time.

Getting Started

Each example within the examples/ directory has its own detailed README.md with specific setup and execution instructions.

Future Plans / Roadmap

This project establishes a foundation for exploring advanced multi-agent systems. Future iterations and new examples aim to explore:

• A2A SDK: Create an example with the officoal A2A Python SDK.
• Asynchronous A2A/MCP: Implement non-blocking calls, leveraging A2A streaming or push notifications, especially for long-running tool operations within Specialist Agent -> MCP Server interactions using the official a2a-sdk.
• Parallel A2A Delegation: Modify Host Agents to delegate tasks to multiple Specialist Agents concurrently (e.g., fetch stock price AND company news simultaneously).
• MCP over HTTP/SSE: Explore MCP Tool Servers that use HTTP/SSE transport (instead of just stdio), requiring security considerations like authentication middleware.
• Advanced State Management: Utilize ADK's persistent SessionService options (e.g., Database, Vertex AI) and explore robust state sharing patterns between agents.
• Error Handling & Resilience: Implement more comprehensive error handling, retries, and fallback mechanisms across all communication layers in various examples.
• Additional Specialist Agents & Tools: Expand the system with more diverse agents (e.g., NewsAgent, PortfolioAgent, DocumentAnalysisAgent) and corresponding MCP Tool Servers or direct API tool integrations.
• ADK Evaluation: Integrate ADK's evaluation framework (adk eval) to measure performance and accuracy of different agent setups.
• UI Enhancements: Improve frontend UIs to better visualize multi-agent activity, tool calls, and potentially handle more complex inputs/outputs.
• Agent Framework Interoperability: Investigate and demonstrate patterns for integrating agents built with other frameworks (e.g., LangChain, CrewAI) into an A2A/MCP ecosystem, potentially via wrappers.
• Security Considerations: Further explore secure authentication and authorization mechanisms for A2A and MCP communication, especially for MCP over HTTP.

📜 License

This project is licensed under the Apache License 2.0. See the LICENSE file.

🤝 Contributing & Disclaimer

This is a personal project by Heiko Hotz to explore Gemini capabilities and agent interoperability. Suggestions and feedback are welcome via Issues or Pull Requests.

This project is developed independently and does not reflect the views or efforts of Google.