Okay, I'll be honest - I think AI companies suck at naming in general. When OpenAI came out with their newest, breakthrough models that introduced advanced reasoning, they called it - drumroll please o1. When they added voice, a completely breakthrough technology foundationally changing how we work with LLMs, the amazingly creative name of "advanced" voice mode was given to it.
Anthropic, IMHO has actually done the best job naming their models, there's a great interview that Lex Fridman does with Dario Amodei, the founder of Anthropic that I highly recommend. In it, Dario shares that naming the three primary models - Haiku, Sonnet, Opus directly correlates to the complexity of the models with Haiku being the most simple and Opus, not surprisingly being the most complex. Pretty neat right?
Yesterday I'd say Anthropic broke their naming streak with the announcement of Model Context Protocol aka MCP. But all is forgiven in my book because MCP is actually a massive step forward in how we'll use LLMs so who the heck cares about what it's called. Still, this name makes MCP sound a lot more technical than it needs to be since it's something I think will be used far and wide. If I were to name it I'd probably call it something like "Agent Intelligence" - that has a nice ring to it doesn't it?
So what is MCP and why do I think it's such a big deal? I'll break this down in a digestible way because I think it's something everyone, whether you currently use LLMs regularly or not, should know.
- First off, MCP didn't come into existence yesterday, it was already live, yesterday Anthropic open-sourced it so anyone can use it and run an MCP Server themselves
- Okay - but what is MCP? At a very high-level, MCP is a a way to connect AI assistants (think agents) to a system where data is stored. The easiest example here, or at least my favorite, is, connecting an AI assistant to a Github repo so that an AI agent can use this data
- Three major components were released yesterday:
- The Model Context Protocol specification and SDKs
- Local MCP support in the Claude Desktop Apps
- An open source repo of MCP Servers
Of course an example is always the best way to showcase how something is being used right? So here's a good one that Anthropic shared alongside the announcement yesterday:
"Early adopters like Block and Apollo have integrated MCP into their systems, while development tools companies including Zed, Replit, Codeium, and Sourcegraph are working with MCP to enhance their platforms—enabling AI agents to better retrieve relevant information to further understand the context around a coding task and produce more nuanced and functional code with fewer attempts."
Okay and that's the TLDR; for you on Anthropic's Model Context Protocol announcement yesterday. If you want to do a deep dive yourself, here's the first two places I would go: