Teaching Software Engineering with LLM and MCP Integration: From Classroom to Industry Practice
This paper proposes an innovative software engineering education framework that integrates Large Language Models (LLMs) and the Model Context Protocol (MCP) into collaborative teaching and industry partnerships to bridge the gap between academic instruction and real-world engineering practices.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine software engineering education as a cooking school. For years, this school taught students how to chop vegetables, boil water, and follow classic recipes (traditional coding). But the world outside has changed: a super-smart kitchen assistant (the Large Language Model, or LLM) has arrived that can chop, mix, and even invent new dishes instantly.
However, there's a problem. If you just hand a student this assistant, they might get confused. The assistant might forget what it was doing halfway through, or it might try to use a blender when you asked for a whisk. It's like having a genius sous-chef who speaks a different language and doesn't know where the spices are kept.
This paper proposes a new way to teach cooking that solves this chaos by introducing a universal remote control called the Model Context Protocol (MCP).
Here is the breakdown of their approach in everyday terms:
The Problem: The "Genius but Clueless" Assistant
Currently, students learn to code with AI tools, but these tools often work in isolation. They are like having a brilliant chef who can't talk to the oven, the fridge, or the other chefs. In the real world, companies need AI that can seamlessly connect to different tools and remember the whole recipe from start to finish. But because schools haven't taught students how to use the "universal remote" (MCP) to manage these connections, graduates are struggling to fit into modern tech jobs.
The Solution: The "MCP-Core, LLM-Assisted" Kitchen
The authors (a team from City University of Hong Kong and HSBC) designed a 16-week course to fix this. They treat the LLM as the powerful engine and the MCP as the steering wheel and dashboard that keeps everything on track.
The 16-Week Journey:
The First Half (Weeks 1–8): Learning the Rules of the Road.
- Students don't just learn to code; they learn how the "universal remote" works. They study how to set up the connection between the AI and the tools it needs. It's like teaching a driver not just how to press the gas, but how to use the GPS, the cruise control, and the emergency brakes so the car doesn't crash.
- Goal: Students understand that AI needs a structured framework (MCP) to be useful in a real company.
The Second Half (Weeks 9–16): The Real-World Test Drive.
- Students move from the classroom to a "simulated garage" and then to a real "dealership" (industry internships).
- They build projects where they have to make the AI talk to different software tools, manage versions of their code, and solve actual problems.
- Goal: They prove they can take the AI, hook it up to the right tools using MCP, and get a job done without the system falling apart.
How They Measure Success
Instead of just giving a test on "what is a variable," they use a mixed scorecard:
- 70% Hard Skills: Did the code work? Did the AI connect to the tools correctly?
- 30% Soft Skills: Did the student show initiative? Could they collaborate with others?
- They also bring in real engineers from companies to help teach, ensuring the lessons match what is actually happening in the tech industry.
What They Admit They Haven't Solved Yet
The authors are honest about the limits of their current "cooking school":
- One Size Doesn't Fit All: Their plan is built for a specific type of university. It might need tweaking for smaller colleges or different majors.
- The Tech Moves Fast: The "universal remote" (MCP) and the "chef" (LLM) are getting updates constantly. The course needs to keep up with new versions and new types of AI (like those that can see images).
- The Industry Connection: Right now, they rely on specific partner companies for internships. They need to figure out how to make this partnership model easy for any school to copy without it being too expensive or difficult.
The Future Plan
To make this better, they plan to:
- Create different versions of the course for different types of schools.
- Build a system that updates the lessons automatically as the AI technology evolves.
- Track students even after they graduate to see if this training actually helps them get hired and succeed in their careers.
In short: This paper argues that to train the software engineers of the future, we can't just teach them to code. We have to teach them how to drive the AI car safely using the right controls (MCP), so they don't crash when they enter the real world.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.