What I build
Focused AI integrations where the AI has to touch something real: a device, an app, or a workflow.
Custom MCP servers
Give AI assistants like Claude safe, well-defined tools for your device, app, API, or internal system, using the Model Context Protocol.
AI-to-hardware integration
Connect AI to real devices: microcontrollers, handhelds, and USB or serial hardware, including MicroPython boards.
AI for creative software
AI control of DAWs, music tools, and desktop apps, built so the AI works the way a musician or creator actually does.
AI-ready developer tooling
Languages, Claude Code skills, and agent workflows designed so AI writes correct code on the first try.
Shipped work
Open-source projects you can install today. Each one started as a real problem I wanted solved. All projects →
PicoCalc MCP
The first MicroPython MCP server for a handheld. AI assistants run code and push files to PicoCalc hardware over USB.
PyPI ยท MCP RegistryView project → MCP serverAbleton MCP
Control Ableton Live from Claude with ~35 MCP tools and a 33-personality style system.
PyPIView project → Languagecr8script
An English-shaped scripting language for LLMs, with errors designed to teach both people and AI agents.
PyPIView project →What "AI-native" means here
AI isn't bolted on at the end. The tools are designed for AI from the start: what the model can see, what it can safely do, and how it recovers when something goes wrong.
I also build with AI every day, so I know where it shines and where it needs guardrails. That shows up in the work: clear tool definitions, honest error messages, and docs written for both people and models.
- Tools for the model: MCP servers with small, well-named, well-documented tools
- Real-world testing: built and tested against actual hardware and software, not mocks
- Safe by design: limits on what the AI can reach and do
- Documented: READMEs and setup guides so your team can run it without me
How a project works
- Fit check. Tell me what you want AI to do. If it's a good match for what I build, we talk it through.
- Scope & quote. A written scope, timeline, and price before any work starts.
- Build. Working checkpoints along the way, tested on the real device or app.
- Handoff. Code, docs, and a walkthrough. Private work is yours; open-source work can come with a video.
Want the build featured on the channel? See integration builds. Every project is scoped and quoted individually.
Questions
What is an MCP server?
MCP (Model Context Protocol) is an open standard for giving AI assistants tools. An MCP server exposes actions, like "run this code on the device" or "add a MIDI clip", that an assistant such as Claude can call. It's how an AI goes from talking about your product to actually using it.
Which AI assistants do your builds work with?
My MCP servers are built and tested with Claude. MCP is an open protocol, so other MCP-compatible clients can connect too; if a specific client matters to you, we'll test against it during the build.
Is the work open source?
By default, yes. Most of what I build is published under MIT and often featured in a video as an integration build. Private work is the exception, and we'd talk that through first.
Do you take every project?
No. This runs alongside my day job, so I take on a few projects a year. Fit comes first: I take on work where I can do something genuinely good, usually involving hardware, music or creative tools, or developer tooling. If it's not a match, I'll say so.
What do I need to provide?
A clear goal, access to the product (hardware, API keys, or a test account), and someone who can answer questions about how it's meant to be used.
Have something AI should actually use?
Tell me about the device, app, or workflow. If it's a fit, I'll come back with ideas and a scope.