Mucka Toolbench

Mucka Toolbench

A personal dev cockpit. One Electron window that supervises four parallel Claude Code agents working on the same project in separate git worktrees, plus a fifth "Mucka" PM agent (voice + text) coordinating them.

Built for a 3840×1200 ultrawide. Works on smaller screens but the layout is happiest with horizontal room.

What's in it

• Four agent panels with real xterm.js terminals backed by node-pty — each in its own git worktree, with attention glow + context-window chip + per-column expand/collapse.
• A "Mucka" PM agent in the top banner — voice (ElevenLabs Conv AI) and text (Claude Agent SDK). 30+ tools spanning read (git, Vercel, PRs, memories, roadmap, the cockpit's own living spec) and write (banner status, attention flags, agent control, PR reviews, deploys, cross-agent broadcast).
• A kanban roadmap with markdown tickets + image attachments. The PM agent can create/edit/move/delete cards.
• A live dual preview pane that loads each agent's dev server with device-viewport presets (iPhone, iPad, Android, desktop) and persistent logins.
• A VSCode-style file explorer with right-click new/rename/delete and ⌘-click URL opens in terminals.
• In-app auto-updater via GitHub Releases.

For the full architecture + capability map see [`MUCKA.md`](MUCKA.md). For agent-side conventions when editing the toolbench itself see [`CLAUDE.md`](CLAUDE.md).

Requirements

• macOS (Apple Silicon or Intel). Build scripts are Mac-only; the Electron app itself runs on Linux/Windows but install scripts and signing are macOS-shaped.
• Node 22+.
• The Claude Code CLI installed and logged in (claude login). Text-mode Mucka uses your Claude Code subscription via the Claude Agent SDK — no API key needed.
• Optional — an ElevenLabs account for voice-mode Mucka. Without it, voice mode shows a "configure" state and text mode still works.
• Optional —