S Guide

A beginner-friendly guide to 5 powerful MCP (Model Context Protocol) servers — Firecrawl, Browser Use, Opik, FastAPI-MCP, and GitHub MCP — explaining what they are, why they matter, and how to use them with AI agents. Written for someone with zero prior knowledge of this topic, and structured to go from basic to advanced.

This repository is a personal study/reference guide. It explains public, freely available tools and links to their official sources so you can explore further.

---

Table of Contents

1. What problem does this solve? (The simple analogy)
2. What is MCP, exactly?
3. Why should you care / when would you use it?
4. The 5 MCP servers covered here (includes the bonus MCP-Use library)
5. Try it yourself: a real, working example
6. How to actually start using one (general steps)
7. Glossary (plain-English definitions)
8. Sources
9. About

---

What problem does this solve? (The simple analogy)

Imagine you hire a brilliant personal assistant (that's the AI). This assistant is extremely smart, but they are locked inside a small room with no phone, no internet, and no keys to any filing cabinet. They can only talk to you — they cannot check your email, browse a website, or look something up on GitHub for you.

MCP is like giving that assistant a keyring and a phone. Suddenly they can:

• Open specific doors (tools) you've given them access to
• Fetch real information from the outside world
• Actually do things (not just talk about them)

Each "door" the assistant can now open is powered by something called an MCP server. This guide covers five of the most useful doors people are currently using.

---

What is MCP, exactly?

MCP (Model Context Protocol) is an open, standard way for an AI model to connect to outside tools — things like websites, databases, file systems, or other apps — without a developer having to write custom, one-off code for every single connection.

Before MCP, if you wanted an AI to talk to, say, GitHub, a developer had to write custom "glue code" specific to GitHub's API. If they then wanted the AI to also talk to a database, they'd write another custom connector. This doesn't scale well.

MCP fixes this by acting like a universal adapter (think of a universal phone charger that works with any wall socket in any country). Any AI model that "speaks MCP" can plug into any tool that has an "MCP server" built for it — no custom wrapper code needed each time.

An MCP server is simply the small program that sits between the AI and a specific tool or service, translating the AI's requests into actions that tool understands (and translating the results back).

---

Why should you care / when would you use it?

You'd reach for an MCP server when you want an AI agent to do more than chat — you want it to take real action or fetch real, live information. A few concrete situations:

You want to...	Which MCP server helps
Pull clean, readable text out of a messy website automatically	Firecrawl MCP Server
Have an AI literally click buttons and fill forms in a browser	Browser Use MCP Server
Track and evaluate how well your AI application is performing over time	Opik MCP Server
Turn your own web app (built with FastAPI) into something an AI can control	FastAPI MCP Server
Let an AI read/manage your GitHub repositories and workflows	GitHub MCP Server

If none of your current projects need an AI to reach outside a chat window, you don't need MCP yet — but it's a foundational skill for building "AI agents" (AI systems that take multi-step actions on their own), which is where a lot of AI engineering is heading.

---

The 5 MCP servers covered here

Full detail for each is in its own file inside the docs/ folder — click through for a beginner-to-advanced explanation, what it's for, and how to try it.

1. docs/01-firecrawl-mcp.md — Firecrawl MCP Server: turns entire websites into clean, AI-readable text.
2. docs/02-browser-use-mcp.md — Browser Use MCP Server: lets an AI control a real web browser.
3. docs/03-opik-mcp.md — Opik MCP Server: monitors and evaluates AI application quality.
4. docs/04-fastapi-mcp.md — FastAPI MCP Server: turns your own API into AI-usable tools.
5. docs/05-github-mcp.md — GitHub MCP Server: lets an AI interact with GitHub repos and workflows.

Plus one extra:

6. docs/06-mcp-use-library.md — MCP-Use: an open-source library to connect any AI model to any MCP server.

And a general walkthrough:

7. docs/07-getting-started-steps.md — the common step-by-step process for trying any MCP server yourself.

A note on how this guide is organized: the 5 servers above are real, third-party tools, explained here conceptually with links to their official setup docs. The runnable code in examples/ (below) is original and self-contained — small demos built specifically to teach the underlying MCP pattern without requiring an account or setup for any of the 5 services. Once the pattern clicks, it transfers directly to Firecrawl, GitHub MCP, or any of the others.

---

Try it yourself: a real, working example

Reading about MCP only gets you so far — examples/hello-mcp-server/ is a small, complete MCP server you can actually run on your own computer. It has been built and tested end-to-end, and walks you through three stages:

1. Build a tiny MCP server with 3 simple tools (add numbers, count words, shout text).
2. Test it locally with a plain Python script — no AI, no API key — to prove the server itself works.
3. Connect it to a real LLM (Claude) and watch the AI decide, on its own, which tool to call based on a plain-English question.

Full step-by-step instructions, expected output, and troubleshooting are in examples/hello-mcp-server/README.md.

A second example, examples/tensorflow-mcp-server/, shows the same idea using a free, offline TensorFlow model instead of a paid LLM — useful for understanding that MCP works with any tool behind it, at no cost and with no API key required. Details are in examples/tensorflow-mcp-server/README.md.

A third example, examples/weather-mcp-server/, shows an MCP tool calling a real external API over the internet (a free, no-key weather service) — the other common pattern besides local-only logic. Details are in examples/weather-mcp-server/README.md.

---

How to actually start using one (general steps)

This is the short version — the detailed, beginner-friendly version (with explanations of why each step matters) is in docs/07-getting-started-steps.md.

1. Pick an MCP client — a program that can "hold" MCP connections (examples: Claude Desktop, Cursor, Windsurf, Cline).
2. Pick an MCP server you want to connect (one of the five above, or any other).
3. Install/configure that server (usually one config file entry, sometimes a Docker command).
4. Restart or reload your MCP client so it detects the new server.
5. Ask your AI assistant to use the new tool in plain English — e.g., "scrape this webpage" or "create a GitHub issue."

---

Glossary (plain-English definitions)

• AI Agent — an AI system that can take a sequence of actions on its own (not just answer one question).
• API — a set of rules that lets two pieces of software talk to each other.
• MCP Client — the app or interface where you interact with the AI, and which manages MCP connections (e.g., Claude Desktop, Cursor).
• MCP Server — the small program that connects one specific tool/service to any MCP-compatible AI.
• Repository (repo) — a project folder tracked by Git, usually hosted on a site like GitHub.
• Docker — a tool that packages software so it runs the same way on any computer.

---

Sources

This guide summarizes and explains, in original wording, ideas from the newsletter article "5 Powerful MCP Servers" by Sumanth P, published on the AI Engineering newsletter (April 21, 2025): https://aiengineering.beehiiv.com/p/5-powerful-mcp-servers

Official project links for each tool are listed in their respective docs/ files.

---

About

Maintained by Yasir Shabir, Computer Science academic — built while learning MCP hands-on, and shared in case it helps someone else get started faster.

Contributions, corrections, and questions are welcome — see CONTRIBUTING.md for how to get involved.

---

License

Feel free to use this guide for personal learning. See LICENSE for details.