Google hands A2A to the Agentic AI Foundation, joining MCP
One Linux Foundation body now stewards A2A, MCP, AGENTS.md and goose — vendor-neutral rails for cross-framework agents. Plus: a registry for AI proofs.
Copy markdown
One roof for the whole agent stack
A2A — the protocol agents use to talk to each other — is now a hosted project at the Agentic AI Foundation, the Linux Foundation's home for open agent infrastructure. It joins MCP (agent-to-tools), OpenAI's AGENTS.md, and Block's goose. The rails your agents run on are no longer any single vendor's to change.
Cross-framework agents, no glue code
Neutral governance plus signed agent cards means agents built on different frameworks can discover each other and hand off tasks without bespoke integrations. Over 150 orgs — AWS, Microsoft, Google, Salesforce, SAP, ServiceNow — back A2A, with WeChat, PayPal, and Azure already running it in production.
What actually changed for you
The spec itself isn't new — A2A v1.0 shipped in March with multi-protocol bindings, version negotiation, multi-tenancy, and signed cards. What changed is the risk profile: building on A2A no longer means betting that Google keeps it open, and MCP and A2A now share one steward.
Elsewhere: a preprint server for AI-written proofs
Terry Tao launched Palomar, a registry that mechanically verifies Lean proofs — human- or AI-generated — using type-checking plus LLM semantic matching. If you're building math or theorem-proving agents, it's now a public place to submit and validate results. It's open for submissions, but explicitly not peer review.