MC Toolkit

mc-toolkit

Open-source Claude Code skills — the public cut of Maciej Cielecki's personal toolkit. Generic, reusable tools with a local customization layer so you can adapt them to your machine and language without forking the code.

Packaged as a Claude Code plugin: every skill invokes as mc-toolkit:<name>.

Skills

Skill	What it does
voicememos	Sync Apple Voice Memos into local, speaker-labeled transcripts — fully on-device (mlx-whisper + silero-VAD + pyannote diarization + voiceprint identification), with optional cloud engines for messy audio.

(More moving over from the private toolkit as they're generalized.)

Install

As a user (marketplace):

claude plugin marketplace add cielecki/mc-toolkit
claude plugin install mc-toolkit@mc-toolkit

Marketplace installs are cached and version-pinned — to pull a new release, run claude plugin update mc-toolkit (editing a cached copy does nothing).

For local development (live-edit): clone the repo and symlink the plugin folder into your skills dir — it's then discovered in place, no caching, edits are live next session:

git clone https://github.com/cielecki/mc-toolkit.git
ln -sfn "$PWD/mc-toolkit/plugins/mc-toolkit" ~/.claude/skills/mc-toolkit

Customization — without forking

Skills read every personal setting (paths, language, interpreter locations, labels) through a small resolver, in priority order:

1. $CLAUDE_PLUGIN_OPTION_<KEY> — Claude Code plugin userConfig
2. $<KEY> — environment variable
3. a <KEY>= line in ~/.claude/.env — also where API keys/secrets live
4. config.local.json next to the skill — gitignored, your overrides
5. the skill's built-in default

So the published code stays generic. To adapt a skill, copy its config.example.json to config.local.json and edit:

cd ~/.claude/skills/mc-toolkit/skills/voicememos
cp config.example.json config.local.json   # then edit your paths/language

config.local.json is gitignored — your values never reach the repo. Secrets (HF_TOKEN, ASSEMBLYAI_API_KEY, …) go in ~/.claude/.env, never in the repo.

Privacy & safety

These skills touch personal data (recordings, voiceprints) but the repo ships none of it. The .gitignore hard-blocks runtime output, snapshots, enrolled voiceprints (*.npy biometric data), audio, and any .env/token files. Sensitive audio is processed fully on-device by default; cloud engines are opt-in and clearly marked.

License

MIT © Maciej Cielecki