Spring AI Alibaba Adds Support for the A2A (Agent-to-Agent) Protocol
Key Points
- Support status: Spring AI Alibaba supports the A2A (Agent-to-Agent) protocol, enabling remote, distributed multi-agent collaboration across processes and services, moving beyond in-process limitations.
- Implementation: A2A client support lives in the
spring-ai-alibaba-graph-coremodule, allowing agents to discover and invoke remote agents via AgentCards, leveraging the A2A SDK. - Availability: Introduced after the 1.0 GA release (June 2025); components such as
spring-ai-alibaba-a2a-commonwere available up to version 1.0.0.4 as of September 2025. - Benefits: Enables enterprise-level distributed AI applications—remote multi-agent workflows with automatic request handling and state synchronization—improving scalability and interoperability.
- https://github.com/alibaba/spring-ai-alibaba
- https://java2ai.com/en/docs/
- https://github.com/alibaba/spring-ai-alibaba/pull/2254
- https://mvnrepository.com/artifact/com.alibaba.cloud.ai/spring-ai-alibaba-a2a-common
- https://java2ai.com/blog/spring-ai-alibaba-10-ga-release/
- https://sca.aliyun.com/en/ai/
Overview of Spring AI Alibaba
Spring AI Alibaba is an open-source agentic AI framework built on Spring AI, tailored for Java developers building ChatBots, workflows, and multi-agent systems. It integrates deeply with Alibaba Cloud's Bailian platform for LLMs (e.g., Tongyi Qianwen), RAG, and other AI services, while supporting enterprise tools like Nacos (service discovery) and ARMS (observability). The framework emphasizes graph-based orchestration via Spring AI Alibaba Graph, allowing declarative definitions of nodes, edges, and states. Details: https://github.com/alibaba/spring-ai-alibaba and https://java2ai.com/en/docs/.The 1.0 GA release in June 2025 introduced production-ready features: graph-based multi-agent frameworks, streaming support, human-in-the-loop interventions, and persistent memory. Pre-built nodes (e.g., LlmNode, ToolNode) and workflow export to PlantUML/Mermaid aid debugging and collaboration.
Understanding the A2A Protocol
A2A (Agent-to-Agent) is an open standard proposed by Google in 2024 enabling autonomous interactions between AI agents. Agents discover capabilities via "Agent Cards" (JSON descriptions), delegate tasks, and communicate in a stateless, secure manner using HTTP, JSON-RPC, and OpenAPI-style authentication. It supports multimodal content (text, audio, video) and long-running tasks. In Spring AI Alibaba, A2A complements MCP (Model Context Protocol) for context sharing with external systems—potentially forming an "MCP + A2A" standard combination for AI ecosystems.Integration and Usage
A2A support enables remote agent invocation for distributed services. Developers build remote agents using classes likeA2aRemoteAgent and A2aNode, which handle URL-based AgentCard parsing, request formatting, and response integration into the overall graph state. A remote writer agent can be configured via a simple builder pattern (e.g., A2aRemoteAgent.builder()), keeping API consistency with local agents. In distributed setups, Nacos serves as the registry for dynamic agent discovery and routing.Key milestones:
| Version/Component | Date | Key A2A Features |
|---|---|---|
| 1.0 GA (core framework) | June 2025 | A2A co-building planned; no direct implementation yet |
| PR #2254 (graph-core update) | August 2025 (merged) | A2A client for remote agent interactions; AgentCard parsing, remote invocation |
| 1.0.0.4 (spring-ai-alibaba-a2a-common) | September 2025 | Stable A2A common utilities |
| Starter A2A Server | September 2025 | Server-side support for A2A-enabled agents |
Related dependencies include a2a-java-sdk-client and a2a-java-sdk-server-common. Developers should add the relevant starters to their Maven pom.xml, configure API keys, and define remote agents using builders. Advanced use cases include distributed research agents (e.g., DeepResearch) with web crawling and Python scripting spread across nodes.