How MCP Integrations Supercharge Your AI Agent

What is MCP

MCP — Model Context Protocol — is a standard way for AI tools like Claude Code to talk to your existing apps. Notion, Google Calendar, Gmail, Obsidian, Slack, Linear, whatever. No custom APIs, no webhooks, no code. One command and the AI can read from (and write to) the tools you already use.

That's the technical explanation. Here's the practical one: MCP is what turns an AI agent from a smart guesser into an informed operator.

Without MCP, your agent works from what you tell it. You type "I have a meeting at 2pm with Sarah" and it plans around that. Fine, but you're doing the work of pulling information and feeding it in. You are the integration layer.

With MCP, the agent pulls your actual calendar, reads your real tasks, scans your inbox, and checks your notes — on its own. It doesn't need you to tell it about the 2pm meeting. It already knows. And it knows about the three other meetings you forgot to mention.

The four tools I connected

I connected all four tool categories to my AI Executive Assistant:

Notion (task manager)

This was the first integration and the most obvious one. My tasks live in Notion, so the EA needed to read them for the morning brief and task tracking.

What it unlocked: Real task data instead of me listing what I remembered. The EA pulls every active task with due dates, statuses, and project tags. It catches things I'd forgotten about — overdue items hiding at the bottom of a board, tasks I created two weeks ago and never looked at again.

Before Notion: "What do you need to do today?" and I'd list whatever came to mind. After Notion: "Here are your 14 active tasks. 3 are overdue. Here's my recommended Top 3 based on impact and urgency."

Obsidian (knowledge base)

My daily notes, project notes, and journal all live in Obsidian. Connecting it gave the EA access to something no other tool could provide: how I've been feeling.

What it unlocked: Energy awareness. The EA reads yesterday's daily note and picks up on mood signals, energy levels, what felt draining, what felt good. If I wrote "crashed hard after 3pm" yesterday, this morning's plan shifts deep work earlier. If I journaled about a rough client call, the EA factors that in.

This is the integration that makes energy-based scheduling actually work. Without it, the EA can ask "how are you feeling?" With it, the EA already has a pretty good idea.

Google Calendar

This one surprised me the most. I'll tell that story in a moment.

What it unlocked initially: Schedule awareness. The EA reads today's meetings — team standups, client calls, masterminds — and calculates available hours. My Notion tasks don't include these standing meetings, so without the calendar integration, the EA would plan as if I had a full open day when I actually had four hours of calls.

What it unlocked later: Writing to the calendar. More on that below.

Gmail

The last integration I added.

What it unlocked: Morning inbox scan. The EA checks for urgent emails that might affect today's plan — a client escalation, a deadline change, something that needs a response before noon. Instead of me opening Gmail and getting sucked into a 45-minute email rabbit hole, the EA surfaces only what matters and I deal with it on my terms.

The calendar moment

When I first connected Google Calendar, I was thinking read-only. Pull my meetings, calculate free time, plan around the fixed commitments. Useful but straightforward.

The real power showed up live, by accident.

I was demoing the EA to a client. Showing off the morning brief, how it pulls tasks and builds an energy-optimized plan. He watched me go through it and then asked: "But can you write to the calendar?"

In that moment, I realized — yes, it can. MCP isn't read-only. It has both read and write access. I'd just never thought to use the write side.

So right there, live, I told the EA: "Put this task in my calendar with yellow and with [EA] in the front." And it did it. Flawlessly. Created a calendar event with the right time, the right color, the [EA] prefix so I could distinguish AI-scheduled blocks from real meetings. First try.

I was genuinely surprised. I knew it was technically possible, but I didn't understand how powerful it would be in practice. Before that moment, my daily plan lived only in the terminal — a text output I'd read and then try to remember. After, the plan was visual. On my calendar. Color-coded. Alongside my real meetings.

Since that demo, I've been using calendar writing every single day. The morning brief creates [EA] blocks for my Top 3 tasks, matched to my energy windows. I can see my whole day — meetings in one color, focus blocks in another, admin time in another. The visual representation changed how I relate to the plan. It's not something I read once and forget. It's there, on my calendar, all day.

All of this was possible because MCP gives both read and write access. One protocol, both directions.

The before and after

Here's what the morning brief looked like at different stages:

No integrations (week 1): The EA asks me what's on my plate. I list what I remember. It suggests a plan based on what I told it. I miss things because I forgot to mention them. The plan exists only as text in the chat.

Notion + Obsidian (week 2): The EA pulls my actual tasks and reads yesterday's journal. It knows things I forgot about. It adjusts intensity based on yesterday's energy. But it doesn't know about my meetings, so it sometimes plans deep work during a call I have scheduled.

All four tools (week 3+): The EA pulls tasks from Notion, reads my energy from Obsidian, checks my calendar for meetings and free time, scans Gmail for urgent items, and builds a plan that accounts for all of it. Then it writes the focus blocks to my calendar so the plan is visual and persistent. I review it, adjust if needed, and my entire day is laid out — in the chat and on my calendar.

The jump from "no integrations" to "all four" wasn't gradual. Each tool added a discrete capability. But the compound effect — all four working together — is what made the EA feel like it genuinely manages my day instead of just suggesting things.

How simple it actually is

Connecting a tool is one command:

claude mcp add google-calendar -- npx -y @anthropic-ai/claude-code-mcp-google-calendar

Or just tell Claude Code: "Connect my Google Calendar." It handles the setup.

That's it. No API keys to manage, no OAuth flows to build, no webhooks to configure. The MCP protocol handles authentication and data access. Your agent gets read and write access to the tool, and every skill that references that tool category starts using real data automatically.

The EA is designed to degrade gracefully — if a tool isn't connected, it falls back to asking you or using cached data. So you can connect tools one at a time and the agent works at every stage. You don't need all four to get value. Even one integration (start with your task manager) makes a noticeable difference.

The real insight

Here's the thing I didn't understand until I lived it: an AI agent is only as good as the data it can access.

A brilliant model with no context is just guessing. It can reason well, write well, plan well — but it's planning based on what you remembered to tell it. And you will always forget things. You'll forget the meeting at 3pm. You'll forget the task that's been overdue for a week. You'll forget that yesterday drained you and today should be lighter.

MCP is the bridge between "smart" and "useful." It gives the agent access to your real world — your actual calendar, your actual tasks, your actual inbox, your actual notes. The model doesn't get smarter. It gets informed. And informed beats smart every time.

Related entries

• AI Executive Assistant — the agent these integrations power
• Building a Stateful Agent on Claude Code — the architecture that the MCP layer plugs into
• Lessons Learned Building My First AI Agent — the full build story
• Energy-Optimized Daily Planning Prompt — the morning brief that uses all four integrations