A2A Protocol Hits 150 Backers as Azure and AWS Embed It

One year, three times the backers

The Agent-to-Agent (A2A) Protocol has reached its first anniversary with numbers that make it hard to dismiss as a Google side project. On April 9, 2026, the Linux Foundation, which hosts the project, said more than 150 organizations now support the standard. It also pointed to deep integration across Google, Microsoft and AWS platforms and to live production deployments in several industries.1 Support has roughly tripled since April 2025, when the protocol launched with just over 50 organizations. The backers now include AWS, Cisco, Google, IBM, Microsoft, Salesforce, SAP and ServiceNow.5

The developer numbers follow the same curve. The core repository passed 22,000 GitHub stars, and the SDKs grew from a single Python implementation to five production-ready languages: Python, JavaScript, Java, Go and .NET.2524 By mid-August, one independent catalog counted more than 25,000 stargazers on the repository.6

The headline numbers are solid. The harder question is what "support" and "production" actually mean here. My view is that A2A has clearly become infrastructure at the platform layer, while the evidence for deep, lasting enterprise use is still thinner than the press materials suggest.

How a Google proposal became neutral plumbing

Google introduced A2A at Cloud Next in April 2025 with more than 50 launch partners, including Accenture, Atlassian, LangChain, MongoDB, PayPal, Salesforce, SAP and UiPath.22 Two months later, at Open Source Summit North America on June 23, 2025, Google gave the protocol to the Linux Foundation for vendor-neutral governance.8 The contribution covered the specification, the SDKs and the developer tooling.8

Accounts of the founding lineup differ slightly. The New Stack listed AWS, Cisco, Salesforce, SAP and ServiceNow alongside Google.7 Other reports said seven founding members, adding Microsoft.8 The gap is minor, but it fits a pattern: much of the protocol's history has been reconstructed from press releases rather than reported independently.

The governance story took another step in August 2026, when A2A moved into the Agentic AI Foundation (AAIF). That Linux Foundation body already hosts Anthropic's Model Context Protocol (MCP), Block's goose runtime and OpenAI's AGENTS.md convention.9 Reported dates for the move vary. Forbes said AAIF announced it on August 17.9 Another explainer gives August 20.18 Some earlier developer guides said A2A had been under AAIF since the foundation launched in December 2025.14 That looks premature given the August announcement. Forbes made the key point clearly: Google had already handed over control in 2025, so the August move changed which foundation hosts A2A, not who controls it.9

What v1.0 changed

The Linux Foundation credits much of the past year's momentum to version 1.0, the first stable specification. It added multi-protocol support, multi-tenancy for enterprises, updated security flows and a defined migration path for early adopters.1 One explainer dates the release to March 12, 2026.2 Forbes also places it in March.9

The most important addition is Signed Agent Cards. An Agent Card is a JSON document, published at a well-known URL, that describes what an agent can do and how to reach it. In v1.0 it can be cryptographically signed so other agents can confirm it really came from the domain owner.22 Without signatures, an attacker could publish a fake card and redirect other agents to it, a risk explainers call card forgery.22 Forbes added a caveat: a signed card only proves as much as the recipient's ability to confirm that the signing key belongs to the claimed organization.9

The protocol is built around long-running tasks. Every exchange is tracked as a Task with explicit states such as submitted, working, input-required and completed.22 Agents built on different frameworks, such as LangGraph or CrewAI, can now hand each other sub-tasks and coordinate workflows without sharing internal memory.1 That opacity is deliberate. Vendors can expose their agents to each other without revealing the internals, which helped make A2A acceptable to large software companies.2

The cloud integrations matter most

The strongest evidence that A2A is becoming a default is the cloud integrations. Microsoft built A2A into Azure AI Foundry and Copilot Studio, and AWS added support through Amazon Bedrock AgentCore Runtime.1 The Linux Foundation says these integrations make A2A a default standard for agent systems in the cloud.1 Google had already announced A2A deployment paths on its own cloud in mid-2025, including Agent Engine, Cloud Run and GKE.10

The ecosystem now extends past the three hyperscalers. One explainer reports that agents built with Strands, OpenAI Agents SDK, LangGraph, Google ADK or the Claude Agents SDK can run as A2A servers inside AgentCore. It also says AWS Marketplace now sells third-party A2A servers.2 AAIF cites Huawei, which uses A2A to connect Celia, its operating-system assistant, with in-app agents on HarmonyOS.9 In June 2026, AWS extended its DevOps Agent so custom SRE agents can be reached over both MCP and A2A.

When a protocol ships inside the managed runtimes enterprises already pay for, adopting it costs very little. That matters more than a list of 150 logos.

Where "production" gets fuzzy

The Linux Foundation says A2A is used in supply chain, financial services, insurance and IT operations to coordinate autonomous systems across tools and vendors.26 Some secondary accounts go further and name production deployments at Microsoft, AWS, Salesforce, SAP and ServiceNow.226 Others add Salesforce Agentforce.2 The anniversary release itself names cloud integrations and industries, not specific production customers.

The more skeptical coverage is right to draw that line. Forbes noted that the adoption figures come from the foundation and its vendors, not from independent measurement. It also warned that implementation depth varies: hosting an A2A endpoint is not the same as supporting protocol bindings, version negotiation, multi-tenancy and signed cards.9 One developer guide argued that "150 organizations support this" shows a commitment to interoperability, not proof that A2A is carrying real production load.24 Another analysis said the measure that matters is retention: whether teams keep A2A running after their first serious operational incident.4

I think the skeptics have the better argument on enterprise depth, but they also concede the main point. Even the most critical pieces agree the spec is real, the SDKs work and the adoption numbers are genuine.24 A2A is not hype. Most teams simply don't need it yet.

A2A and MCP do different jobs

The coverage agrees almost completely on one point: A2A and MCP do different jobs. MCP connects an agent to tools and data. A2A connects agents to each other across organizational boundaries.27 Google has been saying this since the 2025 donation.7 Combined with MCP, A2A gives multi-agent systems a common base that works across technology stacks.27

MCP is still the larger ecosystem by a wide margin. AAIF reported 110 million monthly MCP SDK downloads in April 2026.18 MCP is supported across ChatGPT, Claude, Gemini, Microsoft Copilot, VS Code and Cursor.17 Anthropic donated it to AAIF in December 2025, with OpenAI and Block as co-founders.11 Its July 28, 2026 specification made the core protocol stateless and moved long-running tasks into an official extension.19

That last change shows the two protocols moving toward each other. MCP is adding task handling that A2A has had from the start, while A2A depends on MCP for tool access. The A2A roadmap includes an interoperability specification, a consolidated agent registry, and expanded tooling and security guidance.5 With both protocols now under the same foundation, the two communities have a shared place to coordinate. The specifications remain separate, though, so integration work still falls to the teams deploying them.9

Agent payments: early but notable

A2A is also moving into commerce. The Agent Payments Protocol (AP2) adds secure transactions run by agents, and more than 60 payments and financial-services organizations support it.28 AP2 works as an A2A extension, so an agent can check another agent's Agent Card to see whether it can handle payments.22 It is too early to call A2A the backbone of agent commerce. Still, the payments community's interest suggests the protocol is being taken seriously in settings where transactions must be provably authorized.

What enterprises still have to solve

The gaps lie in what the protocol deliberately does not define. A2A requires servers to check authorization, but it leaves the policy model open. Deciding what a partner's agent may touch after a task crosses a company boundary is still up to each enterprise.9 A2A also standardizes the message envelope, not the contents. Two agents can both advertise a "search inventory" skill and still expect completely different inputs.2

The practical guidance from the more grounded coverage is straightforward. A single agent working with its own tools needs MCP. A2A becomes useful when agents from different vendors or teams have to coordinate long-running work across trust boundaries. A2A's first year shows that the cross-vendor layer of agent systems is becoming standardized. Its second year will show whether enterprises actually rely on it.