← writing
AIMar 22, 2025

MCP: The USB-C for Your LLM

Imagine a world where your frontier LLM integrates with third parties as easily as you connect a USB-C cable between devices.

A few months ago, on a fintech side project, I faced the challenge of integrating an agentic system that validated transactions with multiple external instances. We had to build arguments, parameters, and methods from scratch for the agent to consume — generating hours of extra, repetitive development.

That experience made me reflect: just as we leverage the power of an API today, shouldn't there be an interface that lets LLMs communicate with the outside world as naturally as connecting frontend and backend?

Then November 2024 arrived. Anthropic launched the Model Context Protocol (MCP) — a new open-source bidirectional communication standard. MCP connects AI Assistants, Agents, or LLMs with third-party integrations through a standardized transport layer. Their analogy: the USB-C for your LLM.

With over 1.5k active MCP servers — Notion, Stripe, Figma, Salesforce, Hubspot, and more — extending connections becomes much faster and less painful.

The architecture is simple: - An MCP Host contains the MCP Client - The MCP Client communicates with the MCP Server, which handles integration logic and resources - Everything through a transport protocol

However, interacting with these servers requires a client that understands their transport layer. Today, only apps like Claude Desktop, some AI IDEs (Cursor, Windsurf), and new AI chat platforms meet that requirement. Most MCP Servers are only available to run locally.

Is MCP a standard or a trend? Hard to say. But it's proven ideal for cognitive applications where agent control isn't absolute — connecting Figma's MCP Server to Cursor or Claude Desktop to extract styles and generate React components.

The question is what happens in applications where full control of cognitive instances is required — the majority of what the market needs. There, MCP is just an extra layer.

A truly universal layer capable of interacting naturally with the outside world? Nobody has built that yet.

Originally posted on LinkedIn