Demystifying MCPs

As it turns out, MCP servers aren't just for engineers.

Author

Mike Bal
August 05, 2025

Hopefully you caught my conversation on Supra Insider about my “Aha! moment” with MCPs and the insights I got around improving LLM outputs with more structured memory. I told Ben and Marc I would put out a follow up post and it turned into a bit of a monster so I split it up.

This post will cover the MCP landscape, how to get started, where to use them, etc. The next post will dive deeper into RAG (retrieval augmented generation) and why I think every product team should be using a more human model to improve an LLM’s ability to grok context and give you better responses.

The Current MCP Landscape: Beyond Fleur

The MCP ecosystem has exploded since Anthropic open-sourced it in November 2024. As of May 2025, there are over 5,000 active MCP servers listed in public directories, with major adoption from companies like Block, Replit, Sourcegraph, and even (kind of OpenAI — who adopted it in spring 2025 and only via their API.

When I first jumped in, I was lucky to discover Fleur, which is a simple application that you download, connect to Claude, and that gives you 8 pretty powerfull MCP servers you can toggle on and off.

(See more about my side quest)

MCP Client Options

While Fleur was one of the first user-friendly options, the landscape has expanded significantly:

Fleur inside of Claude

Desktop Applications

Here are some of the desktop apps that make it easy to add MCP “tools” to your LLM workflows and agents.

  • Fleur - macOS desktop app with GUI for MCP server management (original from June.so team)

  • Claude Desktop - Native MCP support with growing integration options

  • 5ire - Cross-platform AI assistant with full MCP support (looks amazing!)

  • DeepChat - Privacy-focused desktop assistant with MCP integration

5ire looks incredible. Trying it today!

Development Environments

Cursor makes it easy but caps total active tool use so you have to manage which ones you need in each session.

Web-Based & Hybrid

To be honest, I haven’t tried any of these yet but I’ll at least acknowledge that they exist :)

  • Glama Chat - Multi-modal AI client with MCP gateway support

  • ChatMCP - Web-based MCP client

  • FLUJO - Workflow-builder interface for AI interactions

MCP Server Directories

Once you figure out what app you are going to use, you have to figure out which MCPs are going to be useful. Some of the apps have connectors and “app store” like directories built in while others require you to do a little more setup and configuration (still super easy).

The ecosystem now has multiple comprehensive directories:

Major Directories (by server count)

  1. HiMCP.ai - 4,700+ servers (largest directory)

  2. MCP.so - 3,056+ servers with clean interface

  3. Glama MCP Directory - 1,617+ servers with quality ratings

  4. Cursor Directory - 1,800+ servers, coding-focused

  5. Smithery - 2,211+ servers with install commands

  6. PulseMCP - 1,704+ servers with GitHub links

Specialized Directories

Getting Started: The Non-Technical Setup

Step 1: Choose Your Client

For product managers, I recommend starting with one of these options:

Easiest Start: Fleur + Claude Desktop

  • Download Fleur from fleur.so

  • Install and give it permissions to talk to Claude

  • Open Claude desktop and type "hello fleur"

  • Choose which MCP servers to activate from the GUI

Most Powerful: Cursor (if you can get past the I of an IDE)

  • Built-in MCP marketplace via Cline extension

  • Shared knowledge graphs across coding and product work

  • Allows you to run tools like Claude Code in parallel with the Cursor Agent

  • Better for technical product managers

Step 2: Start with Memory MCP

The Memory MCP is perfect for PMs because it:

  • Requires no API keys or external accounts

  • Runs locally on your machine (privacy-friendly)

  • Lets you build knowledge graphs without technical setup

  • Mirrors the context building and decision making process we do every single day.

  • Will immediately improve your interactions with the AI and the deliverables you work on together.

Key repositories:

Note: there are other MCPs now that will also let you build a knowledge graph, but they are a bit more complicated to set up. If you want a more technical challenge and something overly robust, you can check this out https://github.com/getzep/graphiti

Step 3: Brain Dump to Build Your Knowledge Graph

Start simple. A knowledge graph creates nodes/entities and then relationships between them. This is naturally how humans tend to process and store information so all you really need to do is think out loud here.

My structure tends to be something like

  • Start a new chat

  • tell it we are going to build out the knowledge graph around a specific thing (the product, a feature, the business, the brand, a user segment, etc.)

  • Share my research, docs, etc. and ask it to get started.

  • See what it did and then start talking to it normally and add additional context where it’s helpful.

    • For example: “While X feature is launching as a stand alone right now, we plan on a pretty tight integration with the core product and y feature in the next 3 months.”

Step 4: Using Your Knowledge Graph

Once you have basic entities, you can reference them in any Claude conversation:

  • "Check your knowledge graph for user pain points and review this design mockup"

  • "Based on what you know about my project scope, help me prioritize these feature requests"

  • "Reference my team context and suggest how to communicate this technical constraint"

Real talk: The first time Claude gives you a response that actually understands your project context without you having to explain it again, you'll get it. That's the moment when this stops feeling like extra work and starts feeling like magic.

worth noting, it seems like Claude might have a cached version of the knowledge graph so if you create this new entity and then start a new chat, it can’t always find th enew info right away. So stay in the same thread or give it a bit.

Product Manager-Specific Use Cases

Based on my experience, here are the highest-impact use cases for PMs:

1. Design Review and Feedback

Setup: Create entities for user pain points, PRD requirements, and design principles Usage: Upload design screenshots and ask for review against your established criteria Benefit: Consistent, contextually-aware design feedback that references your actual project goals

 

2. Requirement Validation

Setup: Build knowledge graph with user research, business goals, and technical constraints

Usage: Test new feature ideas or requirement changes against your established context

Benefit: Catch scope creep and ensure new features align with core user needs

Bonus Points: Use one of the MCPs that give the LLM the ability to use the browser and have it test your demo.

3. Stakeholder Communication

Setup: Add entities for different stakeholder groups and their concerns/priorities

Usage: Generate tailored updates or explanations for different audiences or just get feedback to see if your messaging will trigger any alarms.

Benefit: More effective communication that addresses specific stakeholder needs

Pro Tip: I keep a separate entity for each key stakeholder with notes on what they care about most and what kind of feedback they have given. Then I can also give it meeting notes and keep a rolling record so I can ensure I don’t miss anything.

4. Competitive Analysis

Setup: Create entities for competitors, their feature sets, and positioning

Usage: Analyze how new features compare to competitive landscape, or leverage the perspective when planning and prioritizing your roadmap.

 Benefit: Strategic context for feature decisions and positioning

5. User Story Refinement

Setup: Knowledge graph with user personas, pain points, and business metrics

Usage: Generate or refine user stories that connect to actual user research, or use your understanding to have it help analyze larger data sets and identify things you may have missed or that may have been misinterpreted.

Benefit: Stories that are grounded in real user needs rather than assumptions

The ecosystem now offers thousands of MCP servers and I have absolutely NOT used all of them. Please also be aware that there’s a difference between connecting to something locally or your individual account and connecting at the company level. If you are using these for work, make sure to talk to your security team before connecting to company accounts.

The best I could do was pull together the ones that offer immediate value for AI first product teams.

Essential Productivity Servers

  • Memory MCP - Local knowledge graph storage (start here)

  • Linear MCP - Connect to Linear issues, projects, and roadmaps - great for collaborating with design and engineering and keeping updates and knowledge centralized but accessible.

  • GitHub MCP - Access repositories, issues, and documentation. Obviously helpful for engineers but also for PMs who want to develop a deeper understanding of how the code works.

  • Slack MCP - Search team discussions and decisions and maybe even pull them into specific updates or tickets in Linear ;)

Project Management & Planning

Research & Competitive Intelligence

Design & User Experience

Data & Analytics

Communication & Documentation

Why This Matters for the Future of Product Management

MCP represents a fundamental shift in how we'll work with AI tools. Instead of having to become prompt engineers, we can focus on building structured knowledge that amplifies our product thinking.

This is particularly important for generalists (which most PMs are) because it lets us leverage our ability to connect dots across domains without getting bogged down in the technical implementation of those connections. You can pull and save context anywhere to icnreas ethe level of detail and clarity you are able to provide.

As I mentioned in the podcast, every response I get back from Claude with context from MCP is 10x better than before - even with project knowledge. MCPs have also allowed me to work through complex technical issues as “a developer” that I never would have been able to figure out on my own. Not marginally better - dramatically better.

Getting Help and Going Deeper

Resources:

Community:

How to Get Started

Here's a little framework I have been using:

  1. Pick an MCP that feels obviously useful: come up with at least three things that would improve how useful your LLM client/tool is IF it had direct access to the thing. For example, if you use Linear

    1. I can create tickets directly from Claude as I identify opportunities while analyzing customer research.

    2. If my engineers add the Linear MCP, they can automatically summarize every session on the relevant ticket without having to write anything out

    3. I can use the documents and stories I already have in Linear to prototype UI, interactions, or even to generate mock data.

  2. Pick the thing that feels like it has the least friction and do it every day (at least once): The goal here is twofold, build the muscle (and brain) memory so you don’t forget the option is there and to also have a good sense of if it’s actually useful.

  3. If it feels (almost) effortless make it effortless: Update your system prompt with instructions on when to call the specific MCP tool.

    1. For example: “If I ask you to “find an issue” I want you to use your Linear MCP to find the most relevant issue based on the context.”

  4. Move on to the next thing.

Remember: The goal isn't to become a developer. It's to build systems that amplify your product thinking and help you make better decisions faster.

The real power of MCP isn't in the technical setup - it's in giving you a thought partner that truly understands the context of your work.

Have you started using MCPs yet? What have you learned? I'd love to hear about your experience - the good, the bad, and the "why didn't anyone tell me this would happen" moments.

P.S. If you're enjoying this exploration of AI tools for product thinkers, you might also want to check out my thoughts on why technical knowledge matters for PMs. And if you want to see more of this kind of content, hit subscribe on Leading Product - I promise to keep sharing the tools and frameworks that are actually moving the needle.