What is
A2A
?
Most platform initiatives fail the same way: they start too big, get funded like projects instead of products, build for too long without testing with users, and confuse renaming an infrastructure team with building a platform team. Instead, start small, learn fast and scale deliberately.
At a glance
Attribute | Value |
|---|---|
Primary use case | Cross-framework AI agent communication, discovery, and task delegation |
Deployment type | Open protocol implemented through SDKs and agent clients/servers; supports local and cloud-hosted agents |
Pricing model | Free/open source under Apache License 2.0 |
API compatibility | HTTP, JSON-RPC, SSE; complementary to MCP |
Supported integrations | LangGraph, CrewAI, Semantic Kernel, ADK, MCP, custom agent frameworks |
Source type | Open |
Cloud provider | Cloud Agnostic |
Required skills | Programming with a supported SDK, AI agent integration, HTTP and JSON-RPC APIs, agent discovery and task lifecycle management |
AEP branch | Agent Infrastructure |
AEP category | Harness |
AEP subcategory | Capability |
AEP detail | Protocols |
Profile
At the AEP level of Protocols, A2A provides a common communication standard for AI agents built with different frameworks and by different vendors. It enables independent agents to discover one another, delegate tasks, exchange information, and share results without requiring them to expose their internal memory, tools, or proprietary logic.
A2A is a machine-to-machine communication layer, not an agent development kit or a user-facing messaging application. It is designed to complement MCP: MCP standardizes agent-to-tool communication, while A2A standardizes agent-to-agent communication.
Focus
A2A is for developers and teams building agentic applications with frameworks such as LangGraph, CrewAI, Semantic Kernel, ADK, or custom solutions. It addresses the challenge of enabling agents from different frameworks to interoperate and coordinate work, including delegating subtasks and exchanging results in complex workflows.
Background
A2A was originally developed by Google and later donated to the Linux Foundation. The project is maintained by a Technical Steering Committee with representatives from AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP, and ServiceNow. The protocol is licensed under Apache License 2.0 and welcomes community contributions. The supplied research does not provide a founding date or funding details.
Features
Cross-framework interoperability
Connects agents built on different platforms, allowing teams to combine them in composite AI systems. The protocol provides a shared communication standard for agents from frameworks including LangGraph, CrewAI, Semantic Kernel, and custom solutions.
Agent discovery and capability sharing
Lets agents discover other agents and their capabilities, supporting selection of an agent for a task. Agent Cards provide JSON metadata describing an agent's identity, skills, endpoints, and security requirements.
Task delegation and lifecycle management
Supports delegation and coordination of work across agents, including long-running operations. Interactions use stateful tasks with IDs, status updates, and lifecycle management.
Messages and artifacts
Allows agents to exchange conversational context, instructions, and generated results. Messages carry context and instructions, while artifacts represent outputs such as text, files, or data.
Extensible communication
Allows capabilities to be added while keeping the core protocol stable. Extensions and custom bindings follow a tiered promotion process.
SDK and framework ecosystem
Provides official SDKs for Python, JavaScript, Java, C#/.NET, Golang, and Rust, alongside samples for clients, servers, and framework integrations. Luby reported more than 22,000 GitHub stars.
Pros and cons
Pros
Provides a shared standard for communication and collaboration between agents built with different frameworks and by different vendors.
Supports agent discovery, task delegation, status updates, and sharing of results, including for long-running tasks.
Designed to work alongside MCP: A2A handles agent-to-agent communication, while MCP handles agent-to-tool communication.
Uses an open protocol licensed under Apache License 2.0, with contributions welcomed from the community.
Cons
A2A is a communication layer, not an agent development kit; building agentic applications requires a separate framework or development approach.
A2A does not define how an agent communicates with its own sub-agents or invokes tools; those interactions require framework-native primitives or MCP.
A2A is designed for machine-to-machine communication and is not an interactive messaging application for human conversations.
FAQs
Frequently asked questions about A2A
How does A2A work?
A2A lets independent AI agents discover one another, delegate tasks, exchange messages, and share results. Agent Cards provide JSON metadata about an agent's identity, skills, endpoints, and security requirements, while stateful tasks track IDs, status updates, and lifecycle.
Which frameworks and protocols does A2A integrate with?
A2A supports integrations with LangGraph, CrewAI, Semantic Kernel, ADK, MCP, and custom agent frameworks. A2A handles agent-to-agent communication; MCP handles agent-to-tool communication.
How much does A2A cost?
A2A's pricing model is Free/open source under Apache License 2.0.
What skills are needed to implement A2A?
Implementing A2A requires Programming with a supported SDK, AI agent integration, HTTP and JSON-RPC APIs, agent discovery and task lifecycle management.
What are A2A's limitations?
A2A is a machine-to-machine communication layer, not an agent development kit or a protocol for tool calls or communication with an agent's own sub-agents. Those functions require a separate agent framework, framework-native primitives, or MCP.
