Model Context Protocol, or MCP, is no longer a niche developer acronym; it is rapidly becoming the diplomatic language of the agent era. MCP defines a common way for AI systems to connect to tools, data sources, and reusable prompts, and its official documentation explicitly likens it to a “USB-C port for AI applications.” That metaphor matters because standards become powerful only when rivals adopt them. On December 9, 2025, Anthropic donated MCP to the Linux Foundation’s Agentic AI Foundation, which was co-founded by Anthropic, Block, and OpenAI, with support from Google and Microsoft—an unusually clear signal that major firms now see interoperability as strategically necessary, not merely fashionable. (anthropic.com)
Google’s recent moves show how this standard is being folded into real products. At Google I/O 2025, the company said it had added native SDK support for MCP definitions in the Gemini API. Its Interactions API now includes MCP support and remote MCP integration, allowing Gemini to call tools hosted on external servers. Then, on March 17, 2026, Google pushed the idea further: developers can combine built-in tools such as Google Search and Google Maps with custom functions in a single request, while “context circulation” preserves tool outputs across steps so the model can reason over them more fluidly. In other words, Google is not just supporting MCP; it is building the orchestration layer that makes tool use feel continuous rather than improvised. (blog.google)
OpenAI and Microsoft are moving in the same direction from different angles. OpenAI says MCP is “becoming the industry standard” and now supports remote MCP servers in the Responses API; ChatGPT also lets organizations build custom connectors via MCP, while developer mode offers full MCP client support for both read and write tools. Microsoft, meanwhile, declared at Build 2025 that it is rolling out broad first-party MCP support across GitHub, Copilot Studio, Dynamics 365, Azure AI Foundry, Semantic Kernel, and Windows 11, while also contributing to MCP’s authorization spec and registry design. On Windows, Microsoft even describes agent connectors as MCP servers placed in a secure on-device registry with identity and audit trails. (openai.com)
So, will AI agents become a kind of “connected OS”? Possibly—but only in a qualified sense. MCP is giving agents shared ports, shared permissions, and increasingly shared discovery mechanisms. Yet security remains the unresolved plot twist: OpenAI warns about prompt injection and malicious MCP servers, while Microsoft argues that MCP standardizes the execution surface without fully governing it. The future, then, is not simply connection; it is governed connection. If that governance matures, MCP may do for agents what operating systems once did for software: turn isolated capabilities into an ecosystem. (platform.openai.com)










