For the first several years of building with language models, every integration was bespoke. Connecting a model to a filesystem, a database, or an internal service meant writing glue code specific to that model and that tool, code that broke whenever either side changed and had to be rewritten for the next model. The Model Context Protocol, an open standard introduced by Anthropic in late 2024, addresses that fragmentation directly by defining a single common interface between models and the tools and data they use, so that an integration is written once and works across any model that speaks the protocol.
The design borrows from a pattern that already proved itself. Much as the Language Server Protocol allowed any editor to obtain language intelligence from any language server through one common interface, MCP defines a standard way for a model to communicate with any tool or data source. A tool exposes itself as an MCP server, the model connects as a client, and the two exchange a defined set of messages regardless of who built either side. The integration is decoupled from the model, which is the property that makes it durable, because the tool no longer needs to know or care which model is calling it.
An MCP server exposes a small, well defined set of primitives.
| PRIMITIVE | WHAT IT PROVIDES |
|---|---|
| Tools | Actions the model can call, such as sending a request or running a query. |
| Resources | Data the model can read, such as files, records, or documents. |
| Prompts | Reusable templates a server hands to the model for common tasks. |
The significance of the protocol is not any individual connector, it is the network effect that a shared standard creates. A tool that speaks MCP is immediately usable by every client that speaks it, and a client that adopts it gains access to the whole ecosystem of existing servers without writing integration code for each, which is the dynamic that turns a protocol into a platform. That is why adoption spread quickly across editors, assistants, and platforms once the specification was open. MCP does not make a model more capable in isolation, it makes it connectable, and for real work the connection to a system's actual tools and data is where most of the practical value lies, because a model that cannot reach the systems it is meant to help is limited no matter how capable it is on its own. The full specification is open and maintained at modelcontextprotocol.io.