Inspiration
I love LOL and the esports scene, and project chronobreak has always been super interesting to me. This, along with an article I found about engine-heavy architecture within League, inspired me to try and solve a real-world problem League developers faced.
What it does
Uses MCP to edit and test its own configurations of script-and-Lua combinations to try and make the code more engine-heavy, reducing scripting lines.
How we built it
We built it using Python due to its vast library mainly; the rest of the tech stack seemed to follow naturally from that. The model that I selected was Sonnet 5.5 for cost efficiency mixed with performance and speed. I integrated it with a custom MCP server I made in python.
Challenges we ran into
Sandboxing was a challenge, as I wanted the agent to be able to do a lot but also the bare minimum required. Making my own server was hard but I found it to be the most rewarding and mentally engaging part of the whole thing because it is such a widespread and useful part of any stack to know. Agents can be finicky, and testing can be expensive (literally), so every run had to be very worth it. It is also quite a large codebase to navigate by myself.
Accomplishments that we're proud of
The amount that we can cut down on scripting length is unreal. Thats my favorite part of the whole thing: it keeps scripting light.
What we learned
I learned a TON about MCP, AI agents, general Python know-how, engine tradeoffs, and so so so much more.
What's next for Script-to-Engine Agent
Make it more granular in its parsing, and put some better rules on it for dealing with enterprise-grade engines.
