I Built a Product Brain MCP to Stop Carrying Context in My Head
What is an MCP, why Product Managers need one, and how to build yours in a single afternoon with no coding experience.
I am going to tell you something that changed how I work, and I want you to actually try it today.
Between building DraftKit, rethinking the AIAdventChallenge, co-writing posts, publishing twice this newsletter, and shipping new features for my paid audience like the Prompt-led Vault, I started looking for ways to stop losing time on things that should be automatic.
Every morning I was opening a blank chat window and spending the first 15 minutes re-explaining the same context to the AI:
Here is the auth flow.
Here is the decision we made last week.
Here is the Slack thread from Tuesday.
I was not using AI to think. I was using AI to remember my stuff, and those are very different things.
The fix took less than two hours to build. This article is that fix, written so you can build it too, even if you have never opened a Terminal in your life.
You do not need a technical background for this. You need a folder, a Terminal, and about two hours.
But first, let me tell you what this is actually costing you every single day.
The roadmap for today’s deep dive:
The Context Tax: Why carrying knowledge in your head is costing you 40% of your day.
What is an MCP: The five-year-old explanation that actually makes sense.
The Security Layer: What the February 2026 Claude Code vulnerability taught us about safe AI connections.
The Zero-to-One Build: The step-by-step tutorial. One folder. One script. One day.
What Changes After: How your conversations with Claude get smarter the moment this works.
Let’s get into it. 👇
The Context Tax
A study published in Harvard Business Review followed 20 teams across three Fortune 500 companies for five weeks. They found that the average digital worker toggles between apps and websites roughly 1,200 times each day, which adds up to just under four hours per week spent reorienting after each switch.
One thousand, two hundred times. Every day. Just switching windows.
If you are a Product Manager, you feel this differently than most. You start your morning in a Slack thread or Teams. You jump into a 30-minute stakeholder call where the whole direction of the project shifts. Then you open a blank document and try to reconstruct what was just decided.
By the time you start writing, the logic is already leaking.
Research published by the American Psychological Association found that even brief mental blocks from shifting between tasks can cost as much as 40% of someone’s productive time.
That is almost half your day. Gone. Not to actual work. To the act of remembering where you left off.
For years, the PM’s value was that of a librarian. The person who remembered why we chose one architecture over another six months ago.
You carried the context in your head like a heavy backpack. That backpack is now a liability. The best builders I know in 2026 have stopped carrying context. They have started building bridges. And the most powerful bridge you can build in an afternoon is called an MCP.
What Is an MCP?
Bear with me; I’ll try to explain as if I were talking to a five-year-old.
MCP stands for Model Context Protocol. Before your eyes glaze over, here is all you need to know.
Imagine Claude is a brilliant chef. Right now, that chef is in a kitchen with no food. Every time you want help, you have to run to the grocery store (your files), buy the ingredients (copy and paste the text), and carry them back to the kitchen yourself.
An MCP is a bridge between your pantry and the kitchen. When the chef needs an ingredient, they reach out and grab it directly. No running. No copying. No re-explaining the recipe from scratch every single session.
In practical terms: you build a small script that sits on your computer. That script tells Claude:
“Here is a folder full of my product specs, meeting transcripts, and decision logs. When I ask you something, go read from there first. Do not make things up.”
That is it. That is an MCP.
The result is that Claude stops being a chatbot you talk at, and becomes a Product Brain that actually knows your product’s history.
Before We Build an MCP
The Security Conversation We Have to Have
I will not skip this part, because I have seen what happens when people do.
On February 25, 2026, Check Point Research published a report disclosing critical vulnerabilities in Anthropic’s Claude Code. Researchers showed that attackers could achieve remote code execution and steal API credentials simply by having a developer open an untrusted project that contained malicious configuration files.
The attack worked by exploiting Claude Code’s project-level configuration file, .claude/settings.json, which can be modified by anyone with commit access to a repository. By embedding malicious commands inside that file, an attacker could compromise a developer’s system the moment they opened the project, with no additional interaction required.
All reported vulnerabilities have since been fully patched by Anthropic before public disclosure. But the lesson stands, and it directly shapes how we build.
This is exactly why we build our own MCPs instead of downloading random ones from strangers on the internet. We control what goes in. We control what the AI can read. We start with read-only access. The AI can see your specs. It cannot rewrite them, delete them, or touch anything outside the folder you define.
Think of it as inviting someone into your house to read your bookshelf. Not giving them a spare key and admin rights.
For the first 30 days, your bridge is a one-way street. The AI audits. You decide. You type the changes.
What You Need Before You Start
None of this requires experience. Here is the full setup.
1. A folder on your computer: Create one called product-vault. Inside it, drop two or three of your recent PRDs saved as .txt or .md files. If you do not have PRDs, paste in a few paragraphs from a recent meeting summary. That counts.
2. Python installed: Go to python.org and download the latest version. Run the installer. Confirm it worked by opening your Terminal and typing python --version. If you see a version number, you are good.
3. FastMCP installed: FastMCP is the standard Python framework for building MCP servers. FastMCP 1.0 was incorporated into the official MCP Python SDK in 2024 and the standalone project is now downloaded a million times a day. Install it by running this in your Terminal:
pip install fastmcp4. Claude Desktop: Download it at claude.ai/download if you have not already.
That is the full setup.
I Break AI Tools So You Don’t Have To
If you’re building a product and want a Senior PM to find the exact point where your users give up, I’m taking on new clients. I’ll check your architecture and test your logic before your users do. You can book a Product Audit.
Building Your Product Brain Step by Step
Step 1: Create the Bridge Script
Open your Terminal. Navigate to your product-vault folder:
cd ~/Desktop/product-vaultCreate a new file called brain.py using any text editor. Paste in this code exactly:
from fastmcp import FastMCP
import os
# Start the Product Brain
mcp = FastMCP("ProductBrain")
# Define what the AI can do: read a spec file by name
@mcp.tool()
def read_spec(filename: str) -> str:
"""Reads a product spec from the local vault folder."""
vault_path = os.path.join(os.path.dirname(__file__), "product-vault")
file_path = os.path.join(vault_path, f"{filename}.md")
if os.path.exists(file_path):
with open(file_path, "r") as f:
return f.read()
return "File not found. Check the filename and try again."
if __name__ == "__main__":
mcp.run()What does this do? It creates a server called “ProductBrain” with one tool: the ability to read a file from your vault by name. Nothing else. It cannot write to files. It cannot access anything outside the folder you define. That is by design.
Step 2: Tell Claude Desktop Where the Brain Lives
Find this file on your computer:
On Mac
~/Library/Application Support/Claude/claude_desktop_config.jsonOn Windows
%APPDATA%\Claude\claude_desktop_config.jsonIf the file does not exist yet, create it. Open it and paste in this configuration, replacing the path with the actual location of your brain.py file:
{
"mcpServers": {
"product-brain": {
"command": "python",
"args": [
"/Users/yourname/Desktop/product-vault/brain.py"
]
}
}
}To find the exact path, open Terminal, navigate to your folder, and type pwd. Copy that output and add /brain.py to the end.
Save the file. Completely quit and restart Claude Desktop.
Step 3: Confirm It Worked
When Claude Desktop restarts, look for a small hammer icon at the bottom of the chat window. That icon means Claude can now see and use your tools.
Type this to test it:
“Use the product-brain tool to read the file called [your file name without .md].”
If it works, Claude will respond with the actual content of your spec. It just read it directly from your computer.
If the hammer icon does not appear, open Terminal and run python brain.py directly. Any error messages will tell you exactly what is wrong. Troubleshooting guide here.
What Changes When This Works
The moment your specs are connected, your conversations with Claude change in a way that is hard to describe until you feel it.
Instead of this:
“Here is our authentication flow. [Paste 800 words of context.] Based on all that, does this new requirement create a conflict?”
You ask this:
“Read the auth-flow spec from the vault and tell me if the new SSO requirement we discussed today creates a conflict with what we documented in Q1.”
Claude goes to the file. It reads the actual spec. It gives you a real answer grounded in your real documentation, not a guess based on what you copied in five minutes ago.
Your meetings change too. You stop taking frantic notes trying to remember every word. You start narrating key decisions into a transcript file that your Product Brain can reference later.
The context becomes automatic. Your only job becomes intent.
Stop chasing AI certificates. You don’t need another badge to prove you know your stuff. You need to build the thing!
At Cozora, we focus on building the skills you actually use at work. These are the high-paying skills that matter right now. It’s about execution, not theory.
I want to make sure you have the right tools to build, so here is the deal:
Paid subscribers: 50% off (monthly or annual)
Free subscribers: 10% off
Stop collecting credentials and start building systems. 🚀
What to Build Next
Once you are comfortable with reading files, here is how the bridge gets more powerful.
You can add a second tool that searches across all files in your vault for a keyword. You can add a tool that reads your most recent meeting notes and compares them against an existing spec to flag contradictions. You can point the vault at a folder that syncs automatically from Notion or a shared drive.
Each new tool is just another function in your brain.py, decorated with @mcp.tool().
But do not do any of that today. Today the goal is one thing: get the hammer icon to appear and ask Claude one real question about a real file in your vault.
That first moment, when the AI reaches into your actual work and finds what it needs without you having to explain everything from scratch, that is the shift. You stop being the context carrier. You become the logic architect.
The Cheat Sheet
You don’t need to be a DevOps engineer to build this. I’ve distilled the entire setup into a single reference guide so you don’t get lost in the process.
This configuration is built for speed and security. It runs entirely on your machine and gives the AI “read-only” access.
This means it can see your architecture and past decisions, but it can’t delete your hard drive or mess with your files.
Follow this table to ensure you have the right pieces in the right places before you run your first command.
Access this visual for free here.
Stop Paying the Context Tax
You are already paying it every single day. Every time you open a blank chat and spend the first 15 minutes re-explaining your product, you are spending cognitive energy that should go toward the actual problem.
The fix took me two hours. The change was immediate.
Start with one folder. Start with one spec. See what it feels like when the AI already knows what you are talking about. That is the Prompt-Led future, and you do not need to be a developer to get there.
Two things you can do right now:
Create your product-vault folder today. Drop one file in it. That is the hardest step.
Share this with a PM who is still copying and pasting context into every chat window. They will thank you.
One question for you:
What is the one document you re-explain to your AI more than anything else? Drop it in the comments. I will tell you exactly how I would connect it.
And as always, stop guessing. Start building. 🚀
Founder & Builder, Prompt-Led Product
Proof of Work
The builds coming from this community keep getting sharper. A few worth reading this week:
“Your users moved on, your backlog” Building the wrong thing is a silent killer. This article from Manuel Salvatore Martone and Alessandro Magionami explains why traditional backlogs fail when user behavior shifts faster than your sprint. Read this to ensure you’re solving today’s problems, not last month’s tickets. Read it here.
“Why your laptop is your next big automation platform” Stop overcomplicating your stack with expensive cloud tools. This guide from Jonas Braadbaart shows how to turn your local machine into a powerhouse for automation. Read this to reclaim control of your workflow without adding another monthly subscription. Read it here.
“How to create a brand-aligned photo” Generic AI images are easy to spot and easier to ignore. This breakdown from AI Meets Girlboss teaches you how to maintain a consistent visual identity using specific prompts. Read this to make your brand look professional and cohesive without a studio budget. Read it here.
“How to start vibe coding?” Syntax is becoming a commodity while intent is everything. This post from Jenny Ouyang introduces the concept of shipping products by focusing on the “vibe” and logic rather than just lines of code. Read this to understand the future of rapid prototyping. Read it here.
“Save credits on Perplexity Computer Use with these advanced techniques” AI costs add up fast when you’re testing complex automations. This guide from Karo (Product with Attitude) breaks down techniques to keep your Perplexity usage efficient without burning through your budget. Read this to ship more for less. Read it here.
“Mastering Claude for Product Marketing” Generic marketing prompts yield generic results. This deep dive by Richard King shows how to leverage Claude’s reasoning for high-impact positioning and messaging. Read this to stop writing like a bot and start selling like a pro. Read it here.
If you published something this week related to AI-led building, PM workflows, or product leadership, drop the link in the comments. I read everything.









Love how you did it Elena! I strongly believe that the future will be inundated (in a good way) by MCPs 😁
Nice! Ok now I need to do it :)))