Technology RadarTechnology Radar

Agent2Agent Protocol (A2A)

protocolmulti-agentopen-source
This item was not updated in last three versions of the Radar. Should it have appeared in one of the more recent editions, there is a good chance it remains pertinent. However, if the item dates back further, its relevance may have diminished and our current evaluation could vary. Regrettably, our capacity to consistently revisit items from past Radar editions is limited.
Trial

The Agent2Agent (A2A) Protocol is an open standard that enables AI agents built by different vendors, in different frameworks, to discover each other's capabilities and coordinate work — the missing complement to MCP's agent-to-tool communication. Reached v1.0 on March 12, 2026, under Linux Foundation governance with 150+ contributing organisations and early production deployments.

Architecture Deep Dive → A2A Protocol Architecture Breakdown — Agent Card spec, JSON-RPC/gRPC/REST bindings, SSE streaming, task lifecycle model, and how A2A and MCP fit together in a multi-agent stack.

The Problem It Solves

MCP solved agent-to-tool communication: a standard way for an agent to call APIs, read files, query databases.

A2A solves the adjacent problem: agent-to-agent communication. When you have a LangGraph orchestration agent that needs to delegate a sub-task to a specialist CrewAI agent running on a different team's infrastructure, there was no standard way to:

  1. Discover what the remote agent can do
  2. Send it a task
  3. Track the task's lifecycle
  4. Receive the result in a structured form

A2A defines all of this.

How It Works

A2A introduces "Agent Cards" — JSON documents that describe an agent's capabilities, input/output formats, and endpoints. This enables capability discovery without custom integration code.

The four core capabilities:

Capability What It Provides
Capability discovery Agent Cards let agents advertise what they can do
Task management Defined lifecycle states: pending → active → completed / failed
Agent collaboration Context and instructions shared across agent boundaries
UX negotiation Adapts to different interface capabilities

Relationship to MCP

These two standards are complementary, not competing:

                    ┌─────────────────────┐
                    │    Orchestrator      │
                    │    Agent (LangGraph) │
                    └──────┬──────┬───────┘
                  MCP      │  A2A │
            (tools/data)   │      │ (delegate to
                           │      │  other agents)
             ┌─────────────▼──┐  ┌▼────────────────┐
             │  GitHub MCP    │  │  Code Review     │
             │  Server        │  │  Specialist Agent│
             └────────────────┘  └─────────────────┘

Use MCP when an agent needs to call a tool or read data. Use A2A when an agent needs to delegate to another agent.

Why It Moved to Trial

A2A was released by Google in April 2025, absorbed IBM's Agent Communication Protocol (ACP) in August 2025, and reached v1.0.0 on March 12, 2026 under Linux Foundation governance. The ACP merger brought REST-native design principles and IBM's backing; the combined protocol is now the consensus standard for agent-to-agent communication.

It moved from Assess to Trial because:

  • v1.0 is shipped — stable spec with JSON-RPC, gRPC, and REST bindings, multi-tenancy, and signed Agent Cards
  • SDKs in 5 languages — Python, JavaScript, Java, Go, and .NET, all under the a2aproject org (22.8K GitHub stars)
  • Early production deployments exist — Tyson Foods and Gordon Food Service are running collaborative A2A systems for supply chain coordination; Google Cloud and Microsoft Azure are building A2A into their managed agent platforms
  • Framework integration — Spring AI added A2A support in January 2026; Google ADK has native support

It's Trial rather than Adopt because:

  • Production deployments are still few — most of the 150+ organisations are contributing to the spec or evaluating, not running cross-vendor agent traffic in production
  • Known gaps remain: no enforced authentication standard, scalability concerns with point-to-point communication, and compliance blind spots in agent-to-agent data exchange
  • The protocol just hit v1.0 two weeks ago — ecosystem tooling is catching up

Ecosystem

A2A has first-class support or partnership from:

  • Atlassian, Box, Cohere, Intuit, LangChain, MongoDB, PayPal, Salesforce, SAP, ServiceNow — at launch
  • Accenture, BCG, Capgemini, Deloitte — leading system integrators
  • 150+ organisations contributing to the specification (note: contributing to the spec is a signal of interest, not a proxy for production adoption)

v1.0 Highlights (March 12, 2026)

  • Three protocol bindings — JSON-RPC, gRPC, and REST
  • Signed Agent Cards — cryptographically verified capability discovery
  • Multi-tenancy — support for multiple organisations sharing infrastructure
  • Redesigned Part structure — cleaner data model for task inputs/outputs
  • ACP merger — IBM's Agent Communication Protocol merged into A2A in August 2025, bringing REST-native design and the BeeAI community

Further Reading