Concept
MCP (Model Context Protocol)
An open standard that lets AI models connect to external tools and data sources in a consistent way - like a universal plug that makes any AI work with any tool.
An AI model on its own can only work with what is in its conversation: text you type, documents you paste in, images you upload. It cannot access your calendar, query a database, run a piece of code, browse the web, or send an email - unless someone builds a specific integration connecting the model to each of those things individually. That is exactly what teams were doing before MCP existed: building custom one-off connections between AI and every tool they wanted it to use.
MCP, introduced by Anthropic in late 2024, standardises this. It defines a common protocol - a shared language - for how an AI model should communicate with external tools. Any tool that is built to speak MCP can work with any AI model that supports MCP, without custom integration work for each pair. Instead of building a dozen different cables for a dozen different devices, everyone uses the same universal plug.
The practical impact is significant. Before MCP, an AI assistant might be connected to a handful of tools that a specific team had taken the time to build integrations for. With MCP, the ecosystem expands rapidly because the work of building a connection only needs to be done once per tool, not once per tool per AI application. Developers are building MCP connections for everything - databases, APIs, file systems, development tools, communication apps.
For users, this means AI assistants are becoming genuinely more capable. An AI that can check your email, look at your calendar, run a database query, and pull up a relevant document - all within a single conversation, without you switching between applications - is qualitatively different from one that can only discuss things you paste in manually.
MCP is also significant for the broader AI ecosystem because it is an open standard, not a proprietary one owned by a single company. This means any AI developer can build tools that work with it, and any AI model provider can support it. Open standards tend to become infrastructure, and MCP is on a path to becoming the infrastructure layer for how AI connects to the rest of software.
Analogy
Think about what happened with USB. Before USB, every device needed a different cable and a different port - printers, cameras, keyboards, drives all used different connectors. USB created one standard that everything could use, and the ecosystem of compatible devices exploded. MCP is trying to do the same thing for AI and tools.
Real-world example
Cursor, the AI code editor, uses MCP to let an AI read your codebase, run commands in the terminal, check error logs, and look up documentation - all through a standardised interface. Because these connections are built to the MCP standard, they can be reused across different AI models and different applications, rather than being rebuilt from scratch each time.
Why it matters
MCP is rapidly becoming the infrastructure layer for AI tool use. If it achieves the kind of adoption that HTTP achieved for the web, it will determine how AI connects to the rest of software - which is why every major AI lab and developer tooling company is paying attention and building compatibility.
In the news
No recent coverage - check back later.
Related concepts