MCP Fundamentals Workshop

Welcome to the MCP Fundamentals Workshop by learnwithparam.com!

This hands-on workshop teaches you the Model Context Protocol (MCP) — an open protocol that standardizes how AI applications connect to external tools and data sources. You'll go from manual tool integration to building MCP servers and multi-server clients, step by step.

Regional pricing is available for eligible learners, with discounts of up to 60% in supported regions. Start here: https://www.learnwithparam.com/ai-bootcamp

---

Workshop Storyline

You're building a DevTools AI Assistant — an internal tool that helps your engineering team find research papers (arXiv) and look up technical concepts (Wikipedia). You'll start by wiring everything up manually, feel the pain of the M x N problem, then rebuild it with MCP to see everything click.

---

Workshop Structure (10 Steps)

Module 1: The Problem

Step	File	Description
01	01-why-mcp.md	The M x N integration problem and why MCP exists

Module 2: Meet the Protocol

Step	File	Description
02	02-mcp-architecture.md	Host, Client, Server roles + Tools, Resources, Prompts
03	03-transport-protocols.md	STDIO vs SSE vs Streamable HTTP — when to use each

Module 3: Tool Use Without MCP (The Pain)

Step	File	Description
04	04-tool-use-arxiv.py	Manual tool integration for arXiv (notice the boilerplate)
05	05-tool-use-wikipedia.py	Same boilerplate again for Wikipedia (the M x N problem)

Module 4: Build Your First MCP Server

Step	File	Description
06	06-build-mcp-server.py	Build a server from scratch with @mcp.tool()

Module 5: The MCP Client (The Relief)

Step	File	Description
07	07-mcp-client-basics.py	Your first MCP client — CLI, no LLM, STDIO transport
08	08-mcp-client-single.py	Full chat UI client — Streamlit + LLM + SSE transport

Module 6: Beyond Tools

Step	File	Description
09	09-resources-and-prompts.md	Resources (read-only data) and Prompts (pre-built workflows)

Module 7: Multi-Server Architecture

Step	File	Description
10	10-mcp-client-multi.py	Connect to multiple servers — add sources via config

MCP Servers (Referenced by Steps Above)

File	Transport	Features
servers/arxiv-server.py	STDIO	Tools only
servers/wikipedia-server-stdio.py	STDIO	Tools only
servers/wikipedia-server-sse.py	SSE	Tools only
servers/wikipedia-server-full-stdio.py	STDIO	Tools + Resources + Prompts
servers/wikipedia-server-full-sse.py	SSE	Tools + Resources + Prompts
servers/wikipedia-server-full-http.py	Streamable HTTP	Tools + Resources + Prompts

---

Technology Stack

• LLM: LiteLLM (supports Gemini, OpenAI, Anthropic, etc.)
• MCP SDK: FastMCP (Python)
• Chat UI: Streamlit
• APIs: arXiv, Wikipedia
• Environment: UV, Docker

---

Getting Started

Prerequisites

• Python 3.11+
• UV (fast Python package manager)
• An AI API Key (Gemini recommended for its free tier)
• Docker (optional, for containerized setup)
• Node.js (only needed for step 10's filesystem server via npx)

Setup

# 1. Initialize environment and install dependencies
make setup

# 2. Add your API key
# Edit .env and set GOOGLE_API_KEY (or any LiteLLM-supported provider)

Quick Start

# Run the theory (read in order)
# Open 01-why-mcp.md, 02-mcp-architecture.md, 03-transport-protocols.md

# Run the "pain" demos (steps 04-05)
make run-tool-use-arxiv      # Manual arXiv integration
make run-tool-use-wiki       # Manual Wikipedia integration

# Build and test an MCP server (step 06)
# Read 06-build-mcp-server.py, then test with MCP Inspector:
npx @modelcontextprotocol/inspector uv run python 06-build-mcp-server.py

# Run an MCP client (steps 07-08)
uv run python 07-mcp-client-basics.py           # CLI client
make run-server-wiki-sse                         # Start server (terminal 1)
make run-client-single                           # Start client (terminal 2)

# Multi-server (step 10)
make run-client-multi

---

Teaching Guide (For Live Presentations)

Recommended Flow (90-120 minutes)

1. The Problem (10 min): Walk through 01-why-mcp.md. Ask the audience: "How many of you have written custom API integrations? How many times?"

2. Architecture (10 min): 02-mcp-architecture.md + 03-transport-protocols.md. Focus on the three roles and three capabilities.

3. Live Coding: The Pain (20 min): Open 04-tool-use-arxiv.py. Run it. Then open 05-tool-use-wikipedia.py. Point out the copy-pasted boilerplate. Ask: "What happens when we add a 3rd data source?"

4. Live Coding: Build a Server (15 min): Walk through 06-build-mcp-server.py step by step. Show how @mcp.tool() eliminates schemas. Test with MCP Inspector.

5. Live Coding: The Client (15 min): Run 07-mcp-client-basics.py. Show tool discovery. Then run 08-mcp-client-single.py with the SSE server for the full chat experience.

6. Resources & Prompts (10 min): Walk through 09-resources-and-prompts.md. Show the full server variants.

7. Multi-Server (10 min): Show server_config.json, then run 10-mcp-client-multi.py. Add/remove a server entry live.

8. Wrap-Up (10 min): Production considerations, the MCP ecosystem, what to build next.

---

How to Use This Workshop (Self-Paced)

1. Read the theory (steps 01-03) to understand the concepts
2. Run the pain demos (steps 04-05) to feel the M x N problem
3. Build a server (step 06) — read the code, then test it
4. Connect a client (steps 07-08) to see MCP in action
5. Learn Resources & Prompts (step 09) and explore the full server variants
6. Go multi-server (step 10) to see the full power of MCP

---

Make Commands

# Setup
make setup              # Initialize .env, install uv, create venv
make install            # Sync dependencies
make validate           # Validate Python & JSON syntax

# Servers
make run-server-arxiv          # arXiv MCP server (STDIO)
make run-server-wiki-stdio     # Wikipedia MCP server (STDIO)
make run-server-wiki-sse       # Wikipedia MCP server (SSE, port 8000)
make run-server-wiki-full-stdio  # Full Wikipedia server (STDIO)
make run-server-wiki-full-sse    # Full Wikipedia server (SSE)
make run-server-wiki-full-http   # Full Wikipedia server (Streamable HTTP)

# Demos & Clients
make run-tool-use-arxiv    # Step 04: Manual arXiv integration
make run-tool-use-wiki     # Step 05: Manual Wikipedia integration
make run-client-basics     # Step 07: CLI MCP client
make run-client-single     # Step 08: Single-server Streamlit client
make run-client-multi      # Step 10: Multi-server Streamlit client

# Testing & Cleanup
make test               # Run full test suite
make clean              # Remove venv, cache, auto-generated data
make clean-all          # Also remove Docker volumes

---

About

Created with care by learnwithparam.com.