The Problem MCP Solves
USB-C for agents.
Remember when every phone had its own charger? Each device needed its own cable. AI tools had that same problem. Every app needed custom code to connect to every service. Ten apps and twenty services meant two hundred integrations. Then came one standard plug.
- Every phone had its own charger
- Every app needed custom code
- Then came one standard plug
From many to one
The Model Context Protocol, or MCP, was released by Anthropic in late twenty twenty four, and it's since been adopted right across the industry. Now anyone can write one server for a service, and any compatible app can use it. Instead of multiplying apps by services, you add them together. It's why people call it USBC for AI.
Three roles
MCP has three roles. The host is the app that the user sees, such as a desktop assistant, or a code editor. The client lives inside the host, and it connects to one server. The server wraps a tool or a data source. Your agent is usually the host. Other people's servers give it new abilities.
What a server offers
A server can offer three kinds of thing. Tools are functions that the model chooses to call. Resources are data that the app can read, such as files or policies. Prompts are ready made templates, which the user picks. The difference is who's in control. The model, the app, or the user.
| key | value |
|---|---|
| Tools | the model calls them |
| Resources | the app reads them |
| Prompts | the user picks them |
Summary
One standard, with hosts, clients, and servers. Enough theory. Next, you'll build your own server, for Velo Supply.
Key takeaways- MCP is a standard plug
- Host, client, server
- Tools, resources, prompts