Lesson 6 of 30, Module 1: The Whole Idea. You understand what MCP is, why it exists, and you have FELT the problem it solves.
What you can do after this lesson
When several tools each have to reach several applications, the integration cost multiplies (M jobs times N apps). Standardize the connection point once and the cost becomes additive (M servers plus N apps) instead, and the gap between the two grows fast as either number grows. The arithmetic is the reusable idea, not MCP specifically.
Code in this lesson
Every block the lesson shows on screen, in the order it appears.
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/scribe"]
}
}
}
python3 verify_claude_config.py
RESULT: your config worked. Claude is running every server in it.
Use your filesystem tools to list the files in my scribe folder.
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/scribe
"memory": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-memory"]
}
Downloads
- Module 1 checkpoint (scribe-checkpoint-m1.zip): working code as the module leaves it, so a broken session costs you nothing
- The course GitHub repo: every checkpoint, the README and the full lesson list
Prefer to read?
The written version of this part of the course is on Substack: https://genaiunplugged.substack.com/p/why-mcp-was-created-full-course-lesson.
Lesson transcript
One server, two apps
One server. Two applications. And not a single line of code from you.
Hello, and welcome to lesson 6 of the MCP Masterclass. Last lesson you counted 4 pieces of glue and did not enjoy it. Today you watch that number stop growing. And we start with our map, as always.
6 lit boxes, and today both of the app boxes on the left start earning their place.
Last lesson ended with a promise from me. You write the jobs once, and every app picks them up without new code from you. Well, today you watch that happen, and you write no code at all.
What you need first
So what do you need for this one?
You need a second AI app that speaks MCP. Claude Code will do, if you have a paid Claude plan. So will VS Code with Copilot, or Cursor, or Zed, and all 3 of those have a free level.
Pick whichever one you already have. And if you only have Claude Desktop, read along and count with me instead, because the arithmetic still lands.
You also need Node, which you installed in lesson 3.
Installing it in Claude Desktop
So let us install a server into Claude Desktop first.
Open the Claude Desktop settings file from lesson 3, and put this into it.
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/scribe"]
}
}
}
Change that last path to your own scribe folder, and use the full path, starting from the very top of your disk.
Now quit Claude Desktop properly, and open it again. Ask it what files are in your scribe folder, and it will tell you.
Did it actually work
And if you are not sure whether it worked, do not go hunting through menus. There is a checker in your module 1 checkpoint that answers it in one line.
python3 verify_claude_config.py
It shows what you asked Claude to run, whether Claude is awake, and whether it actually started your servers. What you want at the bottom is this.
RESULT: your config worked. Claude is running every server in it.
That checker never prints your config file, and that is on purpose. Some servers need a token to work, and that token sits in there in plain text. So it is not a file to put on a screen, or in a screen share, ever.
One more thing worth noticing. Quit Claude Desktop and your server stops too. Claude started it, Claude owns it, and it dies with the app. That is the pipe we look at in the next lesson.
Now stop and notice what just happened. You wrote no Python at all, and somebody else's MCP server is doing work for you.
When it says the folder is missing
And if it says it cannot find your folder, do not panic and do not go changing your config. Ask it again, and name the tool this time.
Use your filesystem tools to list the files in my scribe folder.
Here is why that matters, and it caught me out for a while. Claude has its own sandbox with its own folders in it. Ask a vague question and it will often look there first, find nothing, and tell you honestly that your folder does not exist.
Having a tool and reaching for a tool are 2 different things. Your server was connected and working the whole time. Nobody had asked it anything.
The same server in a second app
Next, open your second app and add the same server. Every app has a slightly different place to put it, and every one of them takes the same 4 lines.
In Claude Code it is a single command.
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/scribe
Ask that app the same question. It gives you the same answer.
And I want you to sit with that for a moment.
One MCP server. Two apps. Zero new code.
None of last lesson's glue
Go back to the table from lesson 5, the one with 4 pieces of glue in it. Nothing in that table happened here.
You did not write an adapter for Claude Desktop. You did not write a second adapter for the other app. You wrote a total of 4 lines of settings, twice, and the settings were nearly identical.
That is the whole argument for MCP, and you just did it with your own hands.
So let us push it once more, and add a second server to both apps.
"memory": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-memory"]
}
Now you have 2 servers and 2 apps. Under the old way that is 4 pieces of glue you write, and then maintain forever.
Here it is 2 servers, written by other people, plus 2 settings entries by you.
The arithmetic
So here is the arithmetic, plainly.
The old way is M jobs times N apps. Every pair needs its own glue.
The new way is M servers plus N apps. Each server is written once, and each app learns the protocol once.
| Jobs | Apps | Old way | New way |
|---|---|---|---|
| 2 | 2 | 4 | 4 |
| 5 | 4 | 20 | 9 |
| 10 | 5 | 50 | 15 |
With 2 jobs and 2 apps it is 4 either way. That first row is a tie, which is honest and worth saying out loud. Small setups do not need this.
But move to 5 jobs and 4 apps, and 20 pieces of glue become 9. Move to 10 jobs and 5 apps, and 50 become 15. The gap opens fast.
And notice that you are not the only winner here.
The person who wrote that filesystem server wrote it one time. They did not write a Claude Desktop version, and a Cursor version, and a VS Code version.
Their work reached you without either of you doing anything. That is the part that makes an ecosystem grow, and it is why there are thousands of these MCP servers already.
What is still missing
But what is still missing here?
There is a server for reading files, and a server that remembers notes. There is still no server for the thing you actually want.
Your Google Drive drafts are still out of reach. We look at exactly why at the start of module 2, and the answer is more annoying than you expect.
Where the map stands
So where does our map stand at the end of this lesson?
Both app boxes are lit properly now, and so is the pipe between them and the server. You have watched one server feed 2 apps.
The hole in the middle is still a hole. Somebody else's server sits there today, and yours does not exist yet.
Next lesson we will look at who does what inside MCP, and which of the 3 pieces you are going to build.
Bye now, and I will see you in the next lesson.
