Lately, I’ve made quite a bit of progress… Like really.
RAG, SDK, MCP, template, communication, I’ve been touching a bit of everything, sometimes in the right order, sometimes not at all 😉
And most importantly, I started to feel something nice: everything is starting to connect
Bringing some truth into all this
Until now, a good part of my documentation came from my own experiments.
What works, what doesn’t, what “seems to work”… basically real-world experience.
It’s useful, but it has its limits.
So I decided to stop guessing.
I subscribed to Claude.ai, and I had it read a few MO5 classics.
- Thomson MO5 Guide Ed.2 (Cedic Nathan 1985.11)
- Le Manuel Technique du MO5 (Oury, Cedic Nathan 1985)
- Clefs Pour MO5 (Blanchard, Memento PSI 1985)
And then… big cleanup.
Approximations fixed, things I misunderstood, technical details added, and most importantly real references: we move from “it works on my machine” to “it works and we know why”
At the same time, I had it review all the chunks from my RAG.
Same logic, correction, completion, clarification.
Today, the documentation is much more solid, and most importantly, it no longer relies only on my own knowledge 😉
The SDK is getting stronger
With a cleaner documentation, I was able to evolve the SDK (https://github.com/thlg057/sdk_mo5) more confidently.
I added a few important building blocks.
mo5_audio.hto handle sound, and especially to avoid those annoying beeps when pressing a keymo5_music_swito play music, still improvable but workingmo5_joystickto handle a joystick
Important note: audio and music do not work on all emulators. However, they work very well on a real MO5 (at least on mine 😉 ).
I created an example project to see the SDK in action: https://github.com/thlg057/mo5_sdk_examples
Same goal as always, zero installation with Github Codespace, and with three commands it runs.
make setup-codespace, make install, make.
Simple… well, simple-ish 😉
The infrastructure also got a bit of polish
I also got a small reality check:
My template relied on the official lwtools and cmoc URLs, and one day… nothing… Server down… No way to create a project… Blocked because of a URL.
So I took back control and embedded the setups directly into one of my repositories, and now everything goes through GitHub.
It’s more robust, much more.
However, it means I have to track versions myself, but at least it doesn’t break overnight anymore.
Note: if you prefer using the official URLs (http), you just need to modify the URLs in the Makefile, there are dedicated variables at the beginning of the makefile.
First contact with the “outside”
I also tried something I hadn’t done yet: I contacted the MO5 association to present my toolchain.
They suggested I post on their discord to present my work (https://discord.com/channels/281149808931241984/1493228839420301423).
And then… surprise.
Feedback.
People testing.
Experimenting.
Suggesting ideas.
That really feels good 😉
It’s always a bit special to see your project used by others, in a professional context I’m used to it, it’s my job, but here it’s different, it’s personal, there’s a small growing interest, and honestly it feels great, it’s really rewarding 😉
Vibecoding in progress… with a few surprises
I also started a new project: a game.
Yes, another space invaders 😉
But this time, 100% vibecoding with Claude.
I’m currently recording the sessions to make a full video. Of course, nothing ever goes exactly as planned. I bought a microphone for the occasion. And of course, it doesn’t work… I’ll have to deal with it…
The MCP is starting to move
I referenced my MCP in several registries, as explained in Day 18.
And slowly, I’m starting to see some signals.
Repositories being cloned, people passing by. It’s still small, but it’s alive, and that’s already huge (for me 😉).
Conclusion
Everything is not perfect, but clearly, it’s moving forward.
The documentation is stronger, the SDK is becoming more complete, the setup is more robust, and most importantly… there are people on the other side.
And that changes everything.
To be continued 😉