Agentic AI Foundation Advances MCP and A2A Standards With 150-Plus Backers
Updated
Updated · O'Reilly Media · Sep 11
Agentic AI Foundation Advances MCP and A2A Standards With 150-Plus Backers
3 articles · Updated · O'Reilly Media · Sep 11
Summary
Anthropic’s MCP and Google’s A2A are now being developed under the Linux Foundation-backed Agentic AI Foundation, formalizing cross-vendor cooperation on AI interfaces rather than model competition alone.
MCP has moved fastest: its July 28, 2026 spec is the fifth dated revision since November 2024, and maintainers say it already spans 97 million monthly SDK downloads, 10,000 active servers, and clients including ChatGPT, Claude, Gemini and Copilot.
A2A reached v1.0 in April 2026 with more than 150 supporting organizations, signed Agent Cards, five-language SDKs, and availability in Microsoft Copilot Studio, Azure AI Foundry and Amazon Bedrock AgentCore.
The payoff is reusable integrations and a common security boundary—MCP uses OAuth 2.1-based authorization with PKCE and audience checks—but registries, telemetry conventions and delegation across chained agents remain immature.
For engineering teams, the standards promise portability, testing and governance across hosts, yet the foundation’s young stack still requires version pinning, private catalogs and active participation as the protocols keep changing.
As tech giants lock in MCP and A2A standards, are they secretly commoditizing AI models to seize control of the infrastructure?
Will standardizing AI communication create a seamless digital nervous system, or just introduce a massive new attack surface for hackers?
With autonomous agents gaining first-class identities, can new security protocols truly stop malicious prompt injections from triggering catastrophic enterprise breaches?