A2A has moved into a more serious phase. On August 27, the Agent2Agent protocol announced that it had been accepted as a Growth Stage project at the Agentic AI Foundation, the Linux Foundation-backed home for open agent infrastructure.
That matters because A2A is no longer just Google’s answer to agent-to-agent communication. It now sits beside MCP, goose, and AGENTS.md under the same neutral foundation structure. For developers building multi-agent systems, the signal is simple: the industry is starting to separate agent infrastructure from single-vendor platforms.
What Changed
A2A is the protocol for agents talking to other agents. Where MCP connects agents to tools and data sources, A2A defines how agents discover each other, advertise capabilities, negotiate work, and hand off tasks across framework or vendor boundaries.
The new governance move makes A2A a Growth Stage project under the Agentic AI Foundation. In practical terms, that means the project is expected to mature in the open, with a roadmap shaped by a broader community rather than one company.
The official A2A announcement framed the protocol as the horizontal orchestration layer for agents. MCP is the vertical integration layer: agent to database, agent to API, agent to tool. A2A is the horizontal layer: agent to agent, team to team, organization to organization.
That distinction is not academic. Most production agents still start with MCP because they need access to concrete systems first. But once teams run more than one agent, especially across departments or vendors, handoffs become the hard part. Without a shared protocol, every pairing becomes custom integration work.
Why Neutral Governance Matters
Protocol adoption depends on trust. Developers will not build core infrastructure on top of a standard if they believe the standard can be redirected by one vendor’s product strategy.
That is why the Agentic AI Foundation move matters. A2A now shares a governance umbrella with MCP, which makes the agent stack look less like a standards fight and more like layered infrastructure:
- MCP: agent-to-tool and agent-to-data connectivity
- A2A: agent-to-agent coordination
- AGENTS.md: project instructions and agent-readable operating context
- goose and related projects: runtime and developer tooling around agent workflows
No single piece solves the whole stack. Together, they outline what open agent infrastructure is becoming: tools, peers, instructions, identity, and execution environments that can work across products.
For enterprise teams, this reduces platform risk. A team can adopt A2A for coordination without betting everything on Google Cloud. It can use MCP servers from the AgentNDX directory, run agents in a preferred framework, and still design toward a shared protocol surface for inter-agent work.
The Adoption Picture
A2A was already gaining support before the governance move. The protocol reached a stable v1.0 specification, has backing from more than 150 organizations, and is supported across major clouds including Google Cloud, AWS Bedrock AgentCore Runtime, and Microsoft Azure AI Foundry.
The framework layer is also filling in. A2A support now reaches projects like LangGraph, CrewAI, Pydantic AI, AG2, and IBM BeeAI. That matters because protocol adoption usually happens through defaults. Developers rarely adopt standards in the abstract. They adopt the thing their framework makes easy.
Still, A2A is earlier than MCP in day-to-day developer usage. MCP has the advantage of a direct problem: connect an agent to a tool and get value immediately. A2A solves a second-order problem: coordinate multiple agents that may not share a runtime, framework, owner, or trust boundary.
That second problem is real, but fewer teams hit it on day one.
How This Changes the MCP Story
The biggest mistake is treating A2A and MCP as competitors. They are not solving the same problem.
If an agent needs to query Postgres, read a Notion workspace, search documentation, call Stripe, or inspect a GitHub repository, it needs MCP. If one agent needs to delegate a task to another agent running in a different environment, it needs something like A2A.
A real production workflow might use both:
- A planning agent receives a customer request.
- It uses MCP servers to inspect account data, logs, and open tickets.
- It delegates a billing-specific subtask to a finance agent over A2A.
- That finance agent uses its own MCP servers to verify invoices and payment history.
- The first agent receives the result and prepares the final response.
In that workflow, MCP is the tool access layer. A2A is the delegation layer. Confusing the two leads to bad architecture.
What Developers Should Do Now
Most teams should still start with MCP. If your agent cannot reliably access the tools and data it needs, multi-agent orchestration will only add complexity.
But teams building agent platforms, internal agent marketplaces, or cross-department automations should start tracking A2A now. The key design question is not “Do we need A2A today?” It is “Where will agent handoffs become hard once this system grows?”
Watch for three signals:
- Framework defaults. When A2A becomes the default handoff layer in major frameworks, adoption will accelerate.
- Agent discovery. Agent cards and registry-style discovery will determine whether agents can find useful peers without manual wiring.
- Identity and trust. Cross-organization agent work needs signed identities, audit trails, and policy checks before it becomes normal infrastructure.
The Agentic AI Foundation move does not make A2A mandatory. It makes A2A safer to take seriously. The protocol now has a better shot at becoming shared plumbing rather than a vendor-specific feature.
For the agentic web, that is the point. Open agents need open ways to find tools, understand projects, and talk to each other. MCP covers one part of that stack. A2A is trying to cover the next.
FAQ
Q: Does A2A replace MCP? A: No. MCP connects agents to tools and data. A2A connects agents to other agents. Most production systems that need A2A will still use MCP for tool access.
Q: What is the Agentic AI Foundation? A: It is a Linux Foundation-directed home for open agent infrastructure projects. A2A joining it puts the protocol under neutral governance alongside related projects like MCP and AGENTS.md.
Q: Should every developer implement A2A now? A: No. Start with A2A when you need agents to delegate work across frameworks, teams, vendors, or organizations. If you are still building a single-agent workflow, MCP is usually the first protocol to implement.
Q: Why does neutral governance matter for agent protocols? A: Agent protocols may become long-lived infrastructure. Neutral governance gives teams more confidence that a standard will evolve through community needs rather than one vendor’s roadmap.