AI Companies Plan a New Internet: Here’s Their Key
▼ Summary
– The Model Context Protocol (MCP) is a new industry standard, created by Anthropic, that allows AI agents from different companies to access internet tools and data in a standardized way.
– MCP has been rapidly adopted by major tech firms like OpenAI, Google, and Microsoft, and its governance is being transferred to the Linux Foundation to ensure its neutral, open-source growth.
– The protocol functions by authorizing connections between services, enabling AI models to discover and use external tools—like letting Claude interact with Slack—to perform tasks for users.
– Widespread adoption of MCP is seen as crucial for making AI agents faster, more reliable, and profitable, moving beyond the limitations of agents that must navigate websites designed for humans.
– While MCP’s future as a lasting standard is not guaranteed, its development represents a significant step toward more open, interoperable, and secure agentic AI systems.
For the past year and a half, a quiet consensus has formed among the world’s leading artificial intelligence firms. They are converging on a technical approach designed to power the next wave of applications, one that could fundamentally reshape how we interact with digital services. This strategy hinges on creating a common language, allowing AI agents from different companies to seamlessly access information and tools across the entire internet. The goal is to build a functional, interconnected ecosystem of AI agents, a crucial step toward justifying the massive investments poured into the technology. At the heart of this emerging framework is a specification known as the Model Context Protocol, or MCP.
Originally an internal project by two engineers at Anthropic, MCP has seen astonishingly rapid adoption. Major players including OpenAI, Google, and Microsoft have embraced it, with indications that Apple may integrate it into a future version of Siri. The protocol’s momentum is now being formalized. Anthropic is donating MCP to the neutral, non-profit Linux Foundation. Simultaneously, a coalition of companies is launching the Agentic AI Foundation to advance open-source agentic AI, a move expected to accelerate MCP’s growth and establish it as a true industry standard.
So, what does MCP actually do? In simple terms, it provides a standardized way for an AI model to discover and connect to external tools, data sources, and workflows. Imagine asking Claude to send a message to a colleague on Slack. MCP is the underlying system that authorizes the connection, informs Claude that Slack has a message-sending tool available, and allows the two services to communicate to complete the task. Anthropic’s Chief Product Officer, Mike Krieger, describes the effect as a “ping-pong of intelligence” between applications.
This concept might sound familiar to anyone who remembers the rise of Web 2.0, which was built on applications sharing their APIs. MCP represents the next evolution: a new kind of API built specifically for AI-to-application communication. Its creators aspirationally compare it to the ubiquitous USB-C connector, a single standard for countless devices. The widespread adoption of such a protocol is seen as essential for moving users from traditional apps to AI agents, creating the economic ecosystem needed to sustain the industry.
The protocol began humbly. Anthropic engineers David Soria Parra and Justin Spahr-Summers simply wanted to make their company’s chatbot, Claude, more useful for daily work by connecting it to the tools employees actually used. What started as “Claude Connect” gained immediate internal traction during a company hackathon, revealing its broader potential. Released as an open-source project, its adoption exploded externally. Endorsements quickly followed from Microsoft, OpenAI’s Sam Altman, and Google’s Sundar Pichai, accompanied by prominent billboard advertising in San Francisco.
Despite its success, MCP’s close association with Anthropic presented a potential hurdle. While open-source, other companies might hesitate to contribute improvements that could benefit a competitor, and the theoretical possibility remained that Anthropic could one day restrict access. Donating the protocol to the Linux Foundation eliminates these concerns, encouraging broader collaboration. This donation is part of a larger trend, with companies like Block and OpenAI also contributing related agentic AI projects to the foundation, signaling a collective push for standardized communication between AI systems.
Industry observers believe MCP’s timing is perfect. As AI companies grapple with making agent technology profitable and reliable, a common protocol addresses a core weakness: today’s agents often struggle to navigate a web designed for humans. By allowing systems to talk directly to each other in a structured way, MCP promises faster, more accurate, and more successful AI interactions. The long-term vision includes a marketplace where AI agents can use MCP to perform complex tasks on a user’s behalf, like planning an entire trip by booking flights, hotels, and scheduling reminders autonomously.
Adopting any single standard is a significant bet on the future. History is littered with protocols that lost out to alternatives. However, the grassroots, multi-company support for MCP is unusual. The Linux Foundation’s CEO notes he has “never seen anything like this” in terms of organic, rapid demand from organizations wanting to participate.
Looking ahead, the creators hope MCP will become so seamlessly integrated that end-users never need to know it exists. For them, the AI will simply “do its magic.” Furthermore, placing the protocol under neutral stewardship is expected to enhance security, as companies with deep expertise in that area can now confidently contribute improvements. The collective effort to build a secure, reliable standard is widely seen as the key to unlocking a true market for agentic AI, setting the stage for a more interconnected and capable digital world.
(Source: The Verge)
