On 27 October, Apple announced that Keynote, Pages and Numbers will support MCP, so AI agents can read your documents, make edits and run automated actions inside them. If you have been hearing "MCP" everywhere and nodding along, this guide is for you.
MCP in one line
MCP stands for Model Context Protocol. It is an open standard that lets AI apps connect to other tools and data.
The official MCP site describes it as a USB-C port for AI applications. Before USB-C, every phone had its own charger. Now one cable works almost everywhere. MCP does the same for AI. Instead of every AI app building a separate connection to every tool, they all speak one common language.
Why it matters to you (even if you do not code)
On its own, an AI chatbot only knows what you type or paste into it. With MCP, it can:
- Read your calendar and plan your week
- Search your Google Drive or Notion pages
- Create tasks, update a CRM or draft emails inside your actual tools
- Work with design and 3D software
The MCP site gives examples like an agent that uses your Google Calendar and Notion as a personal assistant, and Claude Code building a web app from a Figma design.
Here is one from my own sessions. A learner who runs a logistics business, and joined my AI From Zero program, connected Codex to Blender through MCP and made an animation, entirely on his own, as homework. You can read his full story here. That is the point of MCP. It turns "the AI told me how" into "the AI did it in my tool".
How MCP works, simply
There are three pieces:
- The host. The AI app you use, like Claude, ChatGPT, Cursor or VS Code.
- The MCP server. A small program that offers one tool or data source to AI, like your calendar, a database or Blender.
- The connection between them. The AI app connects to the server, asks "what can you do?", and uses those abilities when your task needs them.
An MCP server can offer three kinds of things:
- Tools. Actions the AI can take, like "create event" or "search files".
- Resources. Information it can read, like files or database records.
- Prompts. Ready-made templates for common tasks.
Some servers run on your own computer. Others run on the internet, and you sign in to them, usually with a login screen like "Sign in with Google", so the AI never sees your password.
MCP vs API vs connectors vs skills
People mix these up all the time, so here is the quick version:
| Term | What it is |
|---|---|
| API | The door one specific app opens for other software. Every app's door is different. |
| MCP | One common plug that any AI app understands. An MCP server often wraps an app's API. |
| Connector (Claude) or App (ChatGPT) | How you switch on an MCP server inside that AI app. |
| Skill | Written instructions that teach the AI how to do a task your way. |
If you want the Claude side in more detail, read my guide to Claude skills, connectors and plugins. A simple way to remember it: MCP gives the AI hands, a skill gives it your method.
Who supports MCP?
Anthropic open-sourced MCP in November 2024. In December 2025 it donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. That matters because it means no single company owns the standard.
The Linux Foundation says MCP has been adopted by Claude, ChatGPT, Gemini, Microsoft Copilot, Cursor and VS Code, among others. And now Apple is adding it to Keynote, Pages and Numbers.
A few practical details:
- Claude: custom connectors using remote MCP servers work on Free, Pro, Max, Team and Enterprise plans. Free users can add one custom connector.
- ChatGPT: connecting your own custom MCP servers is available for Business, Enterprise and Edu workspaces on ChatGPT web.
- Apple iWork: Apple says MCP support needs macOS 27 or later, arriving as a free update for existing Apple Creator Studio subscribers in November.
How to try MCP today
The easiest start does not need any code:
- Open Claude and go to Customize, then Connectors.
- Connect one app you already use, like Google Calendar or Google Drive, from the directory.
- Ask a real question, like "What meetings do I have this week, and which ones need prep?"
- Watch the approval prompts. Claude asks before using a tool. Read what it wants to do before you click Allow.
Once that feels natural, try a custom connector from a company you trust by pasting its remote MCP server address.
Stay safe with MCP
MCP is powerful because it lets AI act inside your tools. That is also the risk. Both Anthropic and OpenAI warn that a malicious or untrusted MCP server can include hidden instructions (prompt injection) that try to make the AI do things you did not ask for.
My rules for learners:
- Only connect servers from companies you trust. No random links from social media.
- Give the smallest permissions you can. If a server asks for more access than it needs, say no.
- Keep approvals on for anything that writes, sends or deletes. Click "Allow always" only for tools you trust to run unsupervised.
- Turn off tools you are not using in that conversation.
- Check your company's AI policy before connecting work accounts.
For more, read my guides on whether it is safe to let an AI agent use your computer and keeping your data private in ChatGPT, Claude and Gemini.
Want to build with MCP, 1:1?
In my 1:1 programs, we connect AI to the tools you actually use at work, from your inbox to your CRM, and build workflows around the work you already do. See the programs at The Arthi AI Collective to find the right fit.