The Agent2Agent Protocol (A2A) is an open protocol for communication and task coordination between independently implemented AI agents. It was introduced by Google in 2025 and is governed as a Linux Foundation project.
A2A is intended for situations where one agent needs to discover another agent's capabilities, delegate work, exchange messages or artifacts, monitor progress, and receive a result without depending on the other agent's internal implementation.
Core concepts in A2A include:
- an Agent Card, which publishes an agent's identity, endpoint, capabilities, skills, and authentication requirements;
- a task, which represents a unit of delegated work with a lifecycle and status;
- messages, which carry communication between participants;
- artifacts, which represent outputs such as documents, data, or generated files; and
- streaming or asynchronous updates for long-running operations.
A2A complements the Model Context Protocol (MCP) but addresses a different boundary. MCP standardizes how an AI application connects to tools, resources, and prompts. A2A standardizes how autonomous or semi-autonomous agents collaborate with one another.
For example, a procurement agent could delegate supplier research to a specialist agent, receive supporting artifacts, and then continue its own workflow. The agents may use different models, frameworks, memory systems, and internal tools as long as they agree on the A2A interface.
Interoperability does not remove the need for security controls. Implementations must authenticate agents, authorize delegated actions, validate artifacts, limit data disclosure, and protect against malicious instructions. Cross-agent messages are external input and may carry prompt injection attacks.
A2A can reduce custom integration work in multi-agent agent orchestration, particularly across organizational or vendor boundaries. However, protocol compatibility alone does not guarantee semantic compatibility, trust, or reliable task completion.
See Google's A2A announcement and the A2A Protocol project for the specification and implementation resources.
The LLM Knowledge Base is a collection of bite-sized explanations for commonly used terms and abbreviations related to Large Language Models and Generative AI.
It's an educational resource that helps you stay up-to-date with the latest developments in AI research and its applications.