A Common Language for AI Tools
Interest in the Model Context Protocol (MCP) has surged since OpenAI announced its support, following Anthropic’s initial introduction in November. The core idea is straightforward: MCP provides a standardized way for AI assistants to connect with external tools and data sources, removing the need for bespoke, per-integration code. Its advocates often compare it to a universal port or, as Microsoft CTO Kevin Scott noted at Build, “akin to HTTP for the internet, where it allows you to do sophisticated things because the protocol itself doesn’t have much of an opinion about the payload that it carries.”
By enabling large language models (LLMs) to access tools in real time, MCP is a significant step toward agentic AI that can act on a user's behalf. The ecosystem of available servers is expanding daily, and companies like Figma have built their own servers to bring design context into developer workflows.
Speed and Scalability
One common question is how an MCP server differs from a direct API call. The distinction lies in efficiency and intelligence. An LLM using a standard API would have to generate and run code for every single interaction—an impractical approach. With MCP, the model is aware of available tools and can invoke them mid-conversation, making the process far faster and more scalable.
This capability is already leading to impressive demonstrations. Developers have built an “automated AI travel agency” with four agents coordinating across Google Maps, Airbnb, Google Calendar, and Weather. Others have created 3D scenes in Blender with just a few sentences, and a major MCP hackathon saw first place go to an observability platform for MCP calls.
Context is Key
However, the value of MCP is not just in the ability to act, but in the quality of the action. As one engineering director put it, MCP “delivers the determinism AI needs for business apps…it can use AI in clever ways rather than being a slop machine.” The protocol itself is only the conduit; the real expertise lies in determining the best context to send through it.
This is where alignment matters. The results from a design tool's MCP server, for instance, are only as good as the alignment between design and code. The server acts as a multiplier for that alignment, not a replacement for it. Far from diminishing craft, this efficiency encourages a more discerning and precise approach to process and output.
Building on a Standard
For MCP to reach its full potential, industry-wide adoption is critical. As Scott argued, ubiquity is the most important trait for a protocol at this layer of the stack. Rather than debating which technical implementation is marginally better, getting a standard that everyone can build on is the greater victory. The lightweight, unopinionated nature of MCP is a key part of its appeal because its usefulness grows with the size of the community that adopts it—build once, use anywhere.
With a standard protocol in place, developers can begin constructing more sophisticated systems. True agency requires autonomy, which in turn demands a triggering mechanism to monitor the environment. MCP clears the path for new types of triggering architecture that have yet to be invented.
MCP gives LLMs something they haven't had before: a common, standardized method to communicate with the external world. As the enthusiasm for an agentic web builds, this foundation is essential for raising the bar on what intelligent, autonomous tools can achieve.
