<aside> 🦥

Sources

Introducing the Model Context Protocol – Anthropic

modelcontextprotocol.io

What is the Model Context Protocol? – Databricks

</aside>

🦥 Sloth's Simple Definition

<aside> 🦥

MCP is a USB-C port for AI agents. It's an open standard that lets any agent plug into any tool or data source through one common protocol, instead of hand-coding a custom integration for every single app.

</aside>

The Problem It Solves

<aside> 🦥

Before MCP, connecting an agent to Gmail, GitHub, and your database meant 3 separate custom integrations, each with its own auth, error handling, and quirks. 10 agents × 10 tools = 100 integrations. 😩

MCP flips that: build it once, and your agent unlocks a whole ecosystem of tools. M + N instead of M × N.

</aside>

How It Works (3 pieces)

<aside> 🦥

• MCP Host – the AI app the user interacts with (Claude Desktop, an AI IDE, your agent).
• MCP Client – lives inside the host and keeps a 1:1 connection to a server.
• MCP Server – a lightweight program that exposes tools/data (e.g. a GitHub server, a Postgres server).

The agent can then discover tools dynamically, call them, and get results back — all in one standard format.

</aside>

Why Programmers Should Care

<aside> 🦥

• Write once, reuse everywhere — your MCP server works with any MCP-compatible agent.
• No more glue code — standardized auth and message format.
• Huge ecosystem — tons of prebuilt servers already exist (GitHub, Slack, filesystems, databases…).
• It's quickly becoming the default way agents connect to the outside world. </aside>

🦥 The One-Liner

<aside> 🦥

Tools are capabilities. MCP is the standard plug that lets agents use those capabilities without custom wiring.

</aside>