A2A, the open standard between agent systems
A different cut of the problem appears once one agent has to work with another, built in a different framework or by a different vendor. That is the cut where I follow Agent2Agent, or A2A, which also sits in the protocols index.
What A2A standardizes
The specification describes A2A as an open standard designed to facilitate communication and interoperability between independent, potentially opaque agent systems. Where agents may have been built in different frameworks, in different languages or by different vendors, A2A offers a common interaction model. The «What is A2A?» page phrases the same cut as an open standard for communication between agents, whatever framework or vendor built them.
The spec's stated aim is to let agents discover each other's capabilities, negotiate interaction modalities (text, files, structured data), manage collaborative tasks and securely exchange information to achieve user goals, without needing access to each other's internal state, memory or tools. Opaque execution is one of the spec's guiding principles, alongside Simple, Enterprise Ready, Async First and Modality Agnostic.
The specification published on the project's site is 1.0.0, shown there as Latest Released Version. Official SDKs, according to the homepage, implement that same A2A v1.0 specification.
History and stewardship
The Google Developers Blog recounts Google's donation to the Linux Foundation and the formation of the Agent2Agent project, with Amazon Web Services, Cisco, Google, Microsoft, Salesforce, SAP and ServiceNow, seeded by the transfer of the specification, the SDKs and the developer tooling. The protocol was later accepted as a Growth Stage project at the Agentic AI Foundation. The official A2A blog describes that move as protection from single-vendor constraints, under the same Linux Foundation umbrella as MCP, goose and AGENTS.md.
The Technical Steering Committee, in the LF AI & Data post on the merger of ACP with A2A, includes representatives from AWS, Cisco, Google, Microsoft, Salesforce, SAP and ServiceNow, and a seat on behalf of IBM. The current homepage points to neutral governance at the Linux Foundation and to the Apache 2.0 license.
IBM's Agent Communication Protocol, or ACP, merged into A2A and ceased to be an active parallel line, according to LF AI & Data.
Discovery, the card and protocol bindings
The spec is organized in three layers: the Canonical Data Model, with Task, Message, Agent Card, Part, Artifact and Extension, in protocol-agnostic definitions; Abstract Operations, including Send Message, Send Streaming Message, Get Task, List Tasks, Cancel Task and Get Agent Card; and Protocol Bindings, with JSON-RPC, gRPC, HTTP/REST and custom bindings.
Discovery starts from the Agent Card, a JSON document that the official Agent Discovery page describes as a digital business card for an A2A Server. The standardized path, in the Well-Known URI style, is https://{agent-server-domain}/.well-known/agent-card.json. The same page records other strategies, such as curated registries (without prescribing a standard API for those registries) and direct configuration or private discovery.
Official SDKs exist for Python, Go, Java, JavaScript, C#/.NET and Rust, and an official Python tutorial walks through AgentSkill, AgentCard, AgentExecutor and serving the card at /.well-known/agent-card.json, with JSON-RPC routes.
An official example
The example below comes from the official specification, at https://a2a-protocol.org/latest/specification/, the JSON manifesto of an «Example Agent».
{
"name": "Example Agent",
"description": "An example agent that echoes messages.",
"supportedInterfaces": [
{
"url": "https://example.com/a2a/v1",
"protocolBinding": "JSONRPC",
"protocolVersion": "1.0"
}
],
"version": "1.0.0",
"capabilities": {
"streaming": false,
"pushNotifications": false,
"extensions": []
},
"defaultInputModes": ["text/plain"],
"defaultOutputModes": ["text/plain"],
"skills": [
{
"id": "echo",
"name": "Echo",
"description": "Echoes the input.",
"tags": ["echo"]
}
]
}
The fit with MCP
A2A itself describes the fit with MCP. On the «A2A and MCP» page, and in compact form on the homepage, MCP connects an agent to tools, APIs and data; A2A connects an agent to other agents; the two are used together. The same page distinguishes flexible, stateful collaboration between agents from the structured integration of capabilities. A2A does not describe AGENTS.md as part of the inter-agent protocol; the relation there is shared governance at the AAIF.
I keep A2A on the inter-agent side of the map, and I follow the project's own wording when MCP enters the same design. The index holds both entries.