Model Context Protocol (MCP) Explained: The Future of AI Tooling

Before MCP, every AI app reinvented the same wheel: how do I let the model call a tool? How do I expose my database to it? Each provider had its own format. Each tool was tightly coupled to one model.

MCP changes that. It's a universal protocol — like HTTP for the AI ecosystem.

What MCP Actually Is

The Model Context Protocol is an open specification, originally introduced by Anthropic, that defines a standard way for AI models to:

• List available tools (/tools/list)
• Call a tool (/tools/call)
• List available resources (/resources/list)
• Read a resource (/resources/read)

That's it. Four endpoints, JSON in, JSON out. Any service can become an "MCP server" by implementing these endpoints, and any AI app can become an "MCP client" by calling them.

Why It's a Big Deal

Before MCP, if you wanted to give Claude access to your Notion workspace, you had to:

1. Read Notion's API docs
2. Write a custom adapter for Claude's tool format
3. Re-write it for GPT-4, then again for Gemini
4. Maintain three versions forever

With MCP, Notion publishes one MCP server. Every model can use it. No adapters, no rewrites.

The Two Roles

• MCP Client — the AI app (Claude, Noomachy, your custom agent). It discovers and calls tools.
• MCP Server — the service exposing tools (your database, your file system, an API wrapper).

A single AI app can connect to dozens of MCP servers simultaneously. Each one adds capabilities.

How Noomachy Uses MCP

Noomachy is MCP-native from day one. Every tool — built-in or third-party — is exposed through the MCP protocol:

• Built-in MCP server — provides web search, file ops, code execution, database query
• Custom MCP servers — users can register any external MCP endpoint in Skills → Add Custom MCP
• Local MCP server — the desktop app ships with a local MCP server that exposes Apple Mail, Notes, Calendar, Reminders, and more

The cloud agent calls all of these through the same protocol. Whether it's hitting a web API or your local Mac, the interface is identical.

Building Your Own MCP Server

A minimal MCP server is just an HTTP service with two endpoints. Here's the shape:

POST /tools/list
{ "tools": [{ "name": "...", "description": "...", "inputSchema": {...} }] }

POST /tools/call
{ "name": "tool_name", "arguments": { ... } }
→ { "content": [{ "type": "text", "text": "..." }] }

That's literally it. Add authentication, rate limiting, and you have a production-ready MCP server.

What Comes Next

MCP is rapidly becoming the de-facto standard. Expect to see:

• Marketplaces of MCP servers (we're building one in the Noomachy Skill Marketplace)
• Local MCP servers for every major desktop app
• Cloud MCP gateways that bundle multiple services
• MCP-to-MCP routing so agents can compose tools from multiple sources

If you're building an AI app today and not using MCP, you're locking yourself into one vendor.

Try It

Noomachy lets you use built-in MCP tools immediately and add your own with one click.

Sign up free →