LLMs have a context problem
Large language models are trained on static data, so questions that reach beyond that training set tend to produce one of two outcomes: a confident hallucination, or a flat "I don't know." For AI agents to be genuinely useful, they need relevant context injected into prompts—whether that's codebase information, repository data, or project documentation.
Getting that context into a prompt typically requires either carefully engineered prompting or bolt-on tooling. GitHub Copilot's @workspace feature, for instance, pulls relevant information from a codebase into a prompt. That kind of auxiliary tooling works, but it gets complicated quickly when you're wiring it across multiple APIs and services.
MCP: a standard for tool connectivity
The Model Context Protocol (MCP), open sourced by Anthropic this past November, aims to solve that fragmentation. It's an open standard for connecting LLMs and AI assistants to data and tools. The core idea is simple: instead of building bespoke integrations for every AI client and every tool, you define connections once against a shared protocol.
MCP adoption started slowly and then accelerated sharply, and that network effect is the point. Because the protocol is model agnostic, any developer can build integrations, and each new adoption makes the standard more valuable for everyone else.
The pattern isn't unprecedented. In 2016, Microsoft released the Language Server Protocol (LSP), which standardized how code editors support programming languages. LSP's success is now largely invisible—language support across editors just works. MCP draws clear inspiration from LSP and applies the same logic to AI tooling. It enables everyone from major tech companies to independent developers to build AI features that work across any client with minimal setup.
Why open standards matter here
LSP's history shows what happens when an open standard gains traction: better tooling, better developer experience, and better end-user outcomes. The same dynamics apply to MCP. An open protocol with broad community backing means no one is locked into a single vendor's integration layer, and innovation compounds as more people build against the same foundation.
Getting started with MCP
For developers looking to build with the protocol, the official MCP documentation is the entry point. The repository of reference implementations shows real-world server examples, and the MCP specification covers the protocol's technical requirements. For those curious about the LSP lineage, Microsoft's language server extension guide explains why that standard was originally built.
GitHub has also released its own official, open source GitHub MCP Server, which wraps GitHub APIs for automation and integration. It's available for public preview, and can be used with agent mode in supported clients.
MCP is still young, but its trajectory mirrors the arc that made LSP indispensable—just for AI tooling instead of code editors.
