Google wants to be the referee, but it's also fielding a team. The company's unveiling of six AI agent protocols this week—MCP, A2A, UCP, AP2, A2UI, and AG-UI—positions Google as the champion of open standards in an era when Anthropic and OpenAI are building increasingly proprietary agent ecosystems. The tension is obvious: can a company with its own commercial interests legitimately position itself as the neutral infrastructure layer for the entire AI industry?
The protocol suite addresses a real developer pain point. Currently, building an AI agent that works across different data sources and communicates with other agents requires custom integration code for every connection. Google's suite promises to eliminate that sprawl by standardizing agent-to-data interactions (via MCP, the Model Context Protocol) and agent-to-agent communication (via A2A). The company demonstrates this with a "kitchen manager" agent example: one system that checks inventory databases, places wholesale orders through UCP, and authorizes payments via AP2—all without bespoke integration code for each function.
But here's where the land grab becomes visible. Google released these protocols with tooling that deeply integrates into its own ecosystem. The Agent Development Kit (ADK) makes implementing A2UI and AG-UI straightforward, and the newly announced Colab MCP Server lets external agents like Claude Code and Gemini CLI connect to Google Colab—but only through Google's defined interfaces. Developers adopting the protocol suite are simultaneously adopting Google's development patterns, tooling, and mental models.
Anthropic's Claude and OpenAI's models have their own SDKs, their own agent frameworks, their own ways of structuring tool use and memory. These work well within their own walls. But developers who want interoperability—to have Claude agents collaborate with Gemini agents, to pull data from any source without per-connection coding—face a fragmented landscape. Google is betting that the appeal of standardization will pull developers toward its protocols, and that proximity to Google's protocols will mean proximity to Google's platform.
The alternative reading: maybe this is simply good infrastructure. Google's six-protocol suite does solve a genuine coordination problem. Standardization, even when led by a commercial player, can create shared benefits. TCP/IP wasn't invented by a neutral body; it emerged from commercial interests too. The protocols themselves could succeed independently of Google's commercial goals if developers find genuine value.
But the competitive dynamics are stark. Gemini Code Assist received substantial updates alongside the protocol news: Agent Mode with Auto Approve for high-velocity workflows, Inline Diff Views for reviewing AI changes, Finish Changes for context-aware code completion, and Outlines for code navigation. These features directly target GitHub Copilot's core developer workflow. Meanwhile, Plan Mode in Gemini CLI adds a read-only environment for analyzing codebases without execution risk—a safety feature that positions Google's CLI as more responsible than alternatives that might execute untrusted code.
The protocols are now live. The question is whether developers will adopt Google's standards because they're open, or because Google made them convenient to implement. In practice, these motivations aren't separable. Every framework comes with opinions baked in. The real test will be whether agents built on Google's protocols can genuinely interoperate with agents built on Anthropic's or OpenAI's tools—or whether "open standard" becomes the branding for Google's preferred way of doing things.