I love standardizations! Over the past couple of weeks, I’ve been testing out different kits and protocols that were recently released for developing agents. While these standardizations are much needed for enterprises, there is currently no comprehensive documentation covering how these kits and protocols interact with each other.
This post summarizes my takes and findings after building a few multiagent systems using these tools, and it’s mostly engineering oriented.
Overview
ADK
The ADK is a new framework introduced at the google next event. This framework makes it easier to build and deploy agents.
The ADK documentation is quite thorough, and you can find templates to quickly build agents in my adk example repository
It’s worth noting that the development kit also works for open source models.
MCP
MCP protocol has been around for some time now, and I plan to write a separate post about it along with the tooling I’ve developed specifically for llms. This protocol outlines how LLMs can access tools when hosted centrally, providing flexibility in adding security layers for these tools.
A2A
A2A protocol was also showcased at the google next event last month. It addresses the interoperability between different agents.
Architecture
Specifically to get started, I will focus on this agentic pattern I used.
Interactions
While each protocol and kit is well documented individually, there’s still a gap when it comes to integrating them seamlessly. To address this, I’ve created a few templates over the past week in my A2A-MCP repository. These templates aim to demonstrate how to use the architecture shown above.
Specifically, the templates create a sample domain agent with tooling from an MCP server through the stdio transport layer, and a master agent that routes requests to the sample agent based on its skills.
Both of these agents are then configured to operate using the A2A protocol.
Looking back at the architecture diagram, you can see the following components at work:
* ADK -> Building Master and Domain Agents
* A2A -> Wrapping the Agents to use A2A Protocol
* MCP -> Configuring Tooling for each agent Based on required capabilitiesTakeaway
With my academic and professional background, I’m excited to see the industry moving towards standardizations and protocols.
I hope this post gives a concise summary to those looking to put these protocols together with concrete examples in my repo. More in the works.. so stay tuned!