The MCP Servers I Use With Claude (and Why)

For the longest time, Claude was the smartest thinking partner I’d ever had and the most useless coworker. It could analyze any problem I described to it. It could draft a presentation outline, write a client email, design a training program. But the moment I said ‘okay, do it,’ I had to copy the answer, paste it into Gmail or Notion or Slack, and watch Claude become a really expensive autocomplete. MCP changed that. Now Claude can read and write inside the tools the work actually lives in, and the difference is the gap between an advisor and an operator.

The one principle: MCP turns Claude from advisor into operator

The version of Claude without MCP can only react to what you paste into the chat. Smart but trapped. MCP gives Claude direct access to the tools the work actually lives in. It can read Notion pages, search your inbox, look at your calendar, scrape a Reddit thread, deploy a Vercel build. Claude stops being the thing you copy text from and starts being the thing that does the work. The cost is some setup time per tool. The benefit is that one conversation can read a meeting brief from Drive, check your calendar, draft a follow up email in Gmail, and remind you to add it to Notion, in the same turn, without you doing anything but asking.

The eight MCPs and what each one actually does

I’ll start with the five I use daily, then the three I only touch for specialized work.

The one that reads and writes my Notion Content Vault

The Content Vault is where every blog idea, draft, and published piece for leadhuman.ai lives. Before MCP, I’d ask Claude to draft something, copy the text into Notion, fill in the metadata fields manually, set the status. Now Claude reads the database directly, finds the draft I’m asking about, edits it in place. The publishing pipeline for the site runs on this connection. A cloud job drafts an article and writes it to the Vault as a Ready draft, and a GitHub Action picks it up to publish. Zero copy paste from me.

If you live in any database or shared workspace (a content calendar, a project tracker, a CRM, a knowledge base), this is the connection that will change your day first.

The one that searches and drafts in my Gmail

I get more email than I want to read carefully. Most is noise. Some needs a real reply. Claude can search my inbox by sender, thread, label, or keyword, summarize what’s actually in there, and draft responses I review before sending. The most useful thing is asking ‘what’s the last thing I heard from this client’ and getting a useful answer in three seconds instead of opening Gmail and digging through six threads.

Same logic for any inbox at any scale. Even with the AI advances of the past two years, inbox triage is still where most knowledge workers lose an hour a day.

The one that sees my calendar

Claude can list my events, find free time, suggest meeting slots, create new events. The most useful version of this is when I’m doing strategy work in another conversation and Claude can check ‘when does that workshop you mentioned actually start’ without me switching tabs or losing the thread.

Planning conversations get sharper when the AI doing the planning can see your real schedule instead of guessing.

The one that reaches into Google Drive

Drive holds everything I share with collaborators and clients at Peak Potential, the consulting practice. Workshop materials, client briefs, course outlines, slide templates. Claude can pull a brief out of Drive, read it, start helping me edit or build a workshop from it without me uploading anything to the chat. For consulting work this is the difference between Claude being a generic helper and Claude being a context aware coworker.

The same setup works for any org’s shared docs. Confidentiality still matters (only docs you’d share with a vendor should be Claude accessible) but the productivity bump is real.

The one that pulls academic research into the conversation

I write a lot of content about L&D and leadership and I’m allergic to claims that sound good but have no grounding. Consensus lets Claude search peer reviewed research while I’m thinking through a topic, find the actual papers behind a popular framework, and pull the evidence I’d otherwise have to dig through Google Scholar to find. The leadhuman.ai content pipeline uses this connection automatically when drafting articles, so every post starts with a research base instead of vibes.

For anyone writing or teaching in a field with active research (L&D, leadership, behavioral science, design, education), this connection turns Claude into a research assistant who actually checks sources.

The three specialized ones

A few more I use for specific work and don’t touch daily:

Apify scrapes Reddit, Twitter, and Google Trends for the AI Scout, the part of my content pipeline that finds pain first topics worth writing about.
Vercel lets Claude check deploy logs and recent deployments so I can debug when the leadhuman.ai site goes weird.
Adobe is connected for creative work (image editing, asset generation) but I lean on it less than the others because most of my creative work happens in the dedicated image pipeline.

Workhorses for specific jobs, not daily drivers.

How they talk to each other

Each MCP is useful on its own, but they get genuinely powerful when you chain them. The pattern I lean on most: a client emails me with a question and I need to surface the relevant context fast before I reply.

I ask Claude to search Gmail for the last few threads with that client, check Notion for any meeting notes or strategy work I’ve done with them, and look at my calendar for when we last met and when we’re meeting next. One prompt, three MCPs, and I’m in the conversation with real context instead of scrolling through six tabs trying to remember what we agreed last quarter.

That’s the part that compounds. Every connected MCP is one less thing I have to dig up myself. Five connections is when the workflow starts feeling like Claude actually knows my work.

Jay at his laptop in a Tokyo office pulling client context from Gmail, Notion, and Calendar at once

Pulling client context from Gmail, Notion, and Calendar in one prompt. Three places that used to require switching between three tabs and remembering what I was looking for in each.

Use this same pattern for your own work

You don’t need eight MCPs to feel the difference. Most people start with one and add more as they figure out where the friction is.

Jay in a Tokyo cafe explaining how to set up an MCP connection to another person across the table

Start with one connection. Add the next when the friction shows up. The full eight stack took me months to build.

A two MCP stack for new managers: Calendar plus Notion. Claude reads your week from Calendar and helps you prep for the meetings that need it most. Notion holds the running notes for each direct report so the prep is grounded in what you’ve actually been working on with them. Two connections, fifteen minutes to set up, and your one on ones get sharper immediately.

A three MCP stack for L&D leads: Drive plus Notion plus Consensus. Drive holds the source materials and client briefs. Notion holds your frameworks and program designs. Consensus grounds whatever you’re designing in real research instead of LinkedIn folklore. Claude does the connective tissue work that usually eats your Friday afternoon.

A two MCP stack for inbox heavy roles: Gmail plus Calendar. Triage your inbox by asking Claude to surface what actually needs a reply, draft the responses, and check your calendar for any scheduling overlaps before you commit. Easily an hour back per day.

A one MCP starter: just Gmail. If you do nothing else, connect Gmail. The first time Claude finds the thread you’ve been hunting for in three seconds, you’ll get it.

What I’d build differently

A few things still rough about MCP, in case you’re about to try it:

The connections sometimes drop and the recovery isn’t graceful. If your Notion auth times out mid conversation, Claude will sometimes confidently tell you it wrote to your database when it didn’t. Always spot check the first time you ask Claude to write something important.

The permissions are coarse. Most MCP servers ask for read or write access to your whole account when you really just want one folder or one label. The security trade off is real and worth thinking about before you connect anything sensitive.

The MCP marketplace is moving fast and there’s no good central place to see what’s available, which ones are actively maintained, and what they actually let you do. You discover them by stumbling across them on Twitter.

And the obvious one: setup is still too technical for most people. The claude.ai connector menu makes it click to connect for hosted MCPs, but local MCPs still require a config file and command line comfort. That’s the next thing that needs to get easier.

Sources:

Model Context Protocol official docs, Anthropic’s central reference for what MCP is and the growing list of servers.
Claude connectors directory, the click to connect MCP marketplace for hosted servers.
How I Built a Council of AI Advisors, the other half of how I get Claude to do real work for me.
How to Use Claude to Build a Personal Knowledge System, the context layer that MCP makes much more powerful.

Part of the Build with AI series on leadhuman.ai.