A2A and MCP: a2ajava - A Framework for Building Cross-Protocol, Cross-Language Servers

[vishalmysore]

Pre-submission Checklist

• This post follows the MCP community guidelines

What would you like to share?

HI All ,

I’ve been building a core framework called a2ajava — a cross-protocol, cross-language library for building agentic applications that speak both MCP (Model Context Protocol) and A2A (Agent-to-Agent Protocol).

🔧 What is a2ajava?

A Java-first library designed to:
✅ Build MCP-compliant servers and tools
✅ Register agents, tools, and resources using clean Java/Kotlin APIs
✅ Support both MCP and A2A natively via the same codebase
✅ Work seamlessly in Spring Boot, Quarkus, or pure Java environments
✅ Enable local or remote execution of AI tools using just a few lines of code

🛠️ Why I built it:
While working with LLMs and agentic systems, I noticed a clear gap in the Java ecosystem for emerging protocols like MCP and A2A. Most existing tooling was heavily oriented toward Node.js, often requiring complex glue code to integrate with LLM-based workflows. Additionally, I found many existing implementations to be either overly complex or too rigid for real-world use. I needed something lightweight, intuitive, and powerful enough to work across both enterprise applications and robotics systems. That’s what led me to create a2ajava — a framework that brings first-class protocol support to Java, allowing any microservice, agent, or backend to seamlessly participate in the agentic ecosystem.

⚙️ Key Features:
🔌 Plug-and-play tool and resource registration
🌐 Support for stdio, HTTP, or socket transport
♻️ Works with Claude, OpenAI, Grok, or any compliant orchestrator
📦 Lightweight, dependency-free core (Spring Boot optional)
🔍 Built-in JSON-RPC, schema validation, and callback mapping

⚔️ Challenges:
Designing a minimal yet extensible schema layer
Supporting multiple execution styles (sync, async, remote)
Debugging protocol alignment across tools from different languages

📘 What I Learned:
Protocol-first architecture is the future of agentic systems
Java is still a powerhouse for scalable agent backends

Clean abstractions unlock powerful integrations with LLMs, RAG, vector search, and more

💡 Tips for Others:
Use a2ajava to expose any Java service as an MCP tool or resource

You don’t need a giant framework — the agent world rewards minimal, modular design

Perfect for AI bots, DevOps agents, robotics, or any logic that needs to be AI-invokable

I’ve open-sourced the library and examples — happy to collaborate or help others integrate it!

Entire Code for the Framework is here

Please feel free to contribute or provide feedback.

Relevant Links

No response