Google has introduced a protocol for user-to-agent and agent-to-agent communication, https://a2aprotocol.ai/
Using this protocol it is possible to create agents that bundle with their own tools, like agents you can create with Rovo (https://www.atlassian.com/software/rovo), but accessible in a standardized way, both programmatically and manually.
As a note: it might be feasible to allow having path variables in paths to agent cards, so that multiple different agents can be hosted in the same Spring app. \ Potentially, this note applies to MCP servers as well, but that's a different issue.
Comment From: ptaillard
Hello, A note to advocate for the integration of the A2A protocol (https://google.github.io/A2A/) within the Spring Framework, given its immense potential to enhance the capabilities and interoperability of agents. As Google has introduced this protocol for seamless user-to-agent and agent-to-agent communication, embracing it within Spring would align with the framework's commitment to flexibility and innovation. Is this something you're considering? Thanks
Comment From: maxiplux
Hello, Guys thank you so much for this amazing project. Please, do you have any plans to add A2A to Spring AI ?
Comment From: busychen
Hello, Guys thank you so much for this amazing project. Please, do you have any plans to add A2A to Spring AI ?
Comment From: aisofhaoisfh
Hello, Guys thank you so much for this amazing project. Please, do you have any plans to add A2A to Spring AI ?
Comment From: csterwa
Thank you all for notifying us of your interest in A2A protocol support in Spring AI. We are currently monitoring the progress on A2A in terms of community, technical capabilities and APIs. As with other integration protocols, we will evaluate adding direct support as it matures and stabilizes.
In the meantime, could some of you provide some context around how you are looking to put A2A in place in your environments? How would you like to use specific protocol capabilities in your Agent apps? What problems do you see this solving for you that are not handled in other ways? What other approaches have you been looking at before A2A was made available? Any more detailed information would be helpful for us to help prioritize.
Comment From: maxiplux
A2A Protocol Support Request for Spring AI
Hi Spring AI Team,
I'm excited to see this discussion about A2A protocol support! As a solo developer already using Spring AI, I'd love to share my perspective on how this could revolutionize agent coordination in the Java ecosystem.
My Vision: Autonomous Workflow Agents
Current Spring AI Setup: - Building workflow agents using Spring Boot microservices architecture - Working with multiple AI providers through Spring AI's abstractions with autonomous multi-representation messaging
The Multi-Agent Scenario I'm Envisioning: I want to create autonomous agents that can independently decide how to represent their responses based on context and creativity. For example:
- A Content Agent determines a topic needs visual explanation → autonomously delegates to a Video Generation Agent
- A Storytelling Agent feels audio would be more impactful → coordinates with an Audio Synthesis Agent
- A Data Analysis Agent realizes insights need interactive visualization → collaborates with a Visualization Agent
The key is agent autonomy - each agent should intelligently choose the most effective medium for their message without human intervention.
Why A2A is Revolutionary for This Vision
Context Sharing Challenges: - Currently building custom solutions for agents to share conversation context and creative intent - Need standardized ways for agents to communicate not just data, but their "reasoning" about why they chose specific representations - Multi-modal message coordination (text → audio → video workflows) requires sophisticated state management
Service Discovery Analogy: A2A could be as transformational for AI agents as service discovery was for microservices. Just like Spring Cloud transformed how services find and communicate with each other, A2A could standardize how AI agents discover capabilities and orchestrate complex, creative workflows.
Specific Protocol Capabilities I Need
Agent Card Discovery:
spring:
ai:
a2a:
agent:
id: "content-workflow-orchestrator"
capabilities: ["text-analysis", "workflow-coordination"]
media-preferences: ["delegates-to-specialists"]
discovery:
enabled: true
registry-type: "kubernetes-native"
Multi-Modal Communication: - Standardized message formats for creative intent sharing - Media type negotiation between agents - Asynchronous workflow coordination with context preservation
Autonomous Decision Making: - Protocol support for agents to query other agents' capabilities - Standardized patterns for creative/contextual decision delegation - Workflow state management across distributed agent interactions
Current Workarounds & Limitations
What I'm Doing Now: - Custom REST APIs between Spring Boot services for agent communication - Manual workflow orchestration with significant boilerplate - Limited context sharing between specialized agents
The Problem: Each agent communication pattern requires custom implementation. There's no standardized way for my workflow agents to autonomously discover and coordinate with media-specialized agents.
Why This Matters for Spring AI's Ecosystem
Competitive Landscape: Python and TypeScript ecosystems already have emerging A2A implementations. Java/Spring Boot developers are falling behind in building sophisticated multi-agent applications.
Spring AI's Opportunity: Bringing A2A to Spring AI could make Java the go-to platform for enterprise multi-agent systems, leveraging Spring's mature ecosystem for production-ready AI agent coordination.
My Interest & Availability
- No immediate business deadline, but eager to experiment and prototype
- Would love to beta test early implementations
- Happy to share learnings and use cases from experiments
Technical Context: - Solo developer with deep Spring ecosystem experience - Focus on creative, autonomous agent workflows
I see A2A as the missing piece that could unlock a new generation of truly intelligent, collaborative AI systems in the Java ecosystem. The autonomous creative decision-making potential is what excites me most about this protocol.
Would love to see this prioritized for Spring AI's roadmap!
Happy to provide feedback or participate in design discussions.
Comment From: markpollack
@ddobrin Any thoughts on your timelines around A2A support?
For the others on the thread, we would be interested in seeing an initial implementation and then have it incubate some time as a project in the https://github.com/spring-ai-community organization.
Thoughts?
Comment From: ddobrin
@markpollack The A2A Java SDK is available here: https://github.com/a2aproject/a2a-java
The SDK implementation aligns at this time with the specification version v0.2.3, and has been outlined in a first blog post by one of the contributors here: article
The SDK repo is being refactored at this time, in order to move the /sdk-jakarta repository to an independent location. The /sdk-quarkus folder will become a reference implementation for TCK validation, and ALL server implementations can be built/versioned/released at independent locations.
I think that an initial implementation can be built in the Spring Community repo. While I have built a server implementation for workshop/demo purposes, I think an open repo is a better approach and it would fill in an existing gap.
@maxiplux I think support for A2A in modern architectures extends beyond accessing remote A2A servers via remote HTTP calls. There is a need to analyze how A2A participates in event-driven architectures as well