On 20 August, Google's Agent2Agent protocol, A2A, formally joined the Agentic AI Foundation at the Linux Foundation, sitting alongside Anthropic's Model Context Protocol, MCP, under the same neutral governance umbrella. Two of the most consequential standards in the entire agentic AI stack are now governed by the same vendor neutral body, backed by more than 250 members including every major cloud provider and model lab. For enterprise architects, this is not a standards trivia item. It is the moment the protocol layer of agentic AI stopped being a bet on any single vendor's roadmap.
Enterprise architects building agentic systems over the past two years have had to make an uncomfortable choice: build against a specific vendor's proprietary agent framework and accept the lock-in, or build against an emerging open standard and accept the risk that the standard might not survive contact with the market. The consolidation of MCP and A2A under the Agentic AI Foundation meaningfully changes that calculus, because it signals both are backed by a governance structure designed to outlast any single contributor's commercial interest.
What MCP and A2A Actually Do, and Why the Distinction Matters
The two protocols solve different problems, and conflating them is a common architectural mistake. MCP standardises the vertical relationship between an agent and the tools, data and systems it needs to do its job, effectively how an agent reaches into an enterprise database, a search index or a local resource. A2A standardises the horizontal relationship between agents, letting one agent publish a structured description of what it can do, an agent card, so another agent can discover it and hand off work without a human brokering the connection. An architecture that only accounts for one of these relationships is only solving half the integration problem agentic systems actually present.
Why Governance Consolidation Changes an Architect's Risk Calculus
Standing up a reference architecture around a protocol that could be abandoned, forked or commercially redirected at a single vendor's discretion has always been a real architectural risk, even when the protocol itself was technically sound. Moving both MCP and A2A under a neutral foundation with more than 250 members, including direct competitors sitting on the same governance body, substantially reduces that risk. It does not eliminate it. Foundations can still deprioritise projects, and open governance does not guarantee long term relevance. But the risk profile of building against these two protocols specifically is now closer to building against HTTP or TLS than to betting on a single company's roadmap.
What This Should Actually Change in a Reference Architecture
The practical shift is less about ripping out existing integrations and more about where new ones get built. Any new agent to tool integration built from here should default to MCP rather than a proprietary connector, unless there is a specific, documented reason not to. Any new agent to agent handoff should be evaluated against A2A's agent card model before a custom discovery mechanism gets built from scratch. Architects who have been deliberately holding off on protocol decisions, waiting to see which standard actually stuck, now have a considerably clearer signal to act on than they did even a few months ago.
The Risk That Does Not Go Away
Protocol governance consolidation solves the standards risk. It does not solve the security and observability risk that sits on top of any agent to agent or agent to tool interaction, regardless of which protocol carries it. An agent card published under A2A still needs to be validated before another agent trusts it. An MCP connection to an enterprise database still needs the same access controls and audit logging any other data access path requires. Architects treating this consolidation as a reason to relax governance around agent interactions are solving the wrong problem.
What This Means for Your Organisation
What we see across the enterprise architecture engagements we run is that organisations are still treating protocol choice as a minor implementation detail, something left to whichever engineering team builds the first agent integration. That is a mistake this consolidation makes more visible, not less. The organisations getting ahead are the ones updating their reference architecture now to default toward MCP and A2A specifically because the governance risk has genuinely changed, not just because the protocols happen to be popular this quarter.
Key Takeaways
- Google's A2A protocol formally joined the Linux Foundation's Agentic AI Foundation on 20 August, sitting alongside Anthropic's MCP under a single neutral governance structure backed by more than 250 members.
- MCP standardises how an agent accesses tools and data, while A2A standardises how agents discover and hand off work to each other. A complete agentic architecture needs both, not one or the other.
- Governance consolidation substantially reduces the vendor lock-in and abandonment risk of building against these two specific protocols, making them a considerably safer long term architectural bet than a proprietary alternative.
- Protocol governance does not replace the security and observability controls agent interactions still require, regardless of which standard carries them.
How Trusenta Can Help
Enterprise Architecture documents where MCP and A2A fit into an organisation's reference architecture, so new agent integrations default to governed standards rather than proprietary connectors by accident.
Enterprise Architecture in the AI Era reassesses existing agent integrations against this protocol consolidation to identify where legacy, vendor specific connectors carry more long term risk than the organisation has accounted for.
Fractional Enterprise Architect gives organisations without a dedicated architecture function the ongoing judgement needed to make protocol decisions as the agentic AI standards landscape continues to consolidate.
Conclusion
Protocol wars rarely end cleanly, and it would be premature to call this one settled for good. What changed on 20 August is still significant: two of the most important standards in the agentic AI stack are now governed by the same neutral, heavily backed foundation, which meaningfully lowers the risk of building a reference architecture around them. Enterprise architects who update their default choices accordingly are making a considerably safer bet than those still waiting for more certainty that, on the current evidence, has already arrived.
