MacCode — Claude Code for the Macintosh SE

MacCode is a native System 7 application for a 68K Macintosh (e.g. a Mac SE) that re-creates the Claude Code terminal UI on classic hardware and relays a Claude Agent SDK session running on a modern Mac. Three layers:

• the SE app — C/Retro68 + the classic Toolbox + MacTCP
• a tiny typed-record wire protocol over plain TCP (no SSL — trusted LAN / the emulator's slirp link)
• a Node/TypeScript proxy — the Claude Agent SDK

You type prompts on the SE; Claude's replies stream back into a scrolling Monaco transcript with an animated ✻ verb line and ● tool lines, tool use is gated by an Allow/Deny dialog, and ! commands run a host shell.

Built on the VibeRetro68 template (Retro68 toolchain + Basilisk II). The toolchain/emulator setup guides live in docs/.

---

Prerequisites

• macOS host. Run make setup once to populate deps/ with the Retro68 cross-toolchain and a preconfigured Basilisk II (System 7.x) with MacTCP over slirp. make setup runs four idempotent steps — brew bundle, make fetch-deps, make build-retro68 (~30–60 min, one-time), make doctor — see docs/RETRO68_SETUP.md and docs/EMULATOR_SETUP.md for detail. deps/ is gitignored; every clone builds its own.
• Node.js for the proxy.
• An authenticated Claude environment for the Agent SDK (the proxy uses your existing Claude credentials).

---

Build the SE app

cmake --build build --target MacCode_APPL

Important: build the MacCode_APPL target, not the bare MacCode. MacCode only links the code; MacCode_APPL runs the Rez/packaging step that produces the runnable build/MacCode.bin / .dsk (and sets the Finder bundle bit for the app icon). If MacCode.bin's timestamp doesn't update after a build, you built the wrong target.

Simpler — the run script builds everything, deploys MacCode.bin into Basilisk II's shared folder, and launches the emulator:

scripts/run-basiliskii.sh MacCode

---

Run the proxy

Start the proxy before launching the app. From the directory you want Claude to work in:

make proxy                 # or: scripts/run-proxy.sh

That serves the current directory as the project on port 4242 (and installs the proxy's dependencies on first run). Flags pass through to override the defaults:

scripts/run-proxy.sh --project ~/some/project --port 4242

Flag	Default	Purpose
--project	directory you run it from	Working dir for the Claude session and for ! shell commands.
--port	4242	TCP port to listen on.
--host	0.0.0.0	Bind address; the default lets the emulator reach the host.
--model	—	Optional model override.
--echo	off	A bring-up echo server. Not for normal use.

The proxy logs timestamped lines to stderr (client connect, <- prompt, -> text, shell:, …).

---

Launch and connect

1. Start the proxy (above).
2. scripts/run-basiliskii.sh MacCode, then open MacCode inside the emulated Mac.

On launch it connects to 10.0.2.2:4242 (the slirp gateway = the host). If the proxy is down it reports "could not connect" but stays responsive (the connect is non-blocking). Use Session ▸ Connect / Disconnect to retry.

---

Using it

Type a prompt and press Return. The ✻ verb line animates while Claude works, ● tool lines show tool calls, and the reply streams into the transcript.

• Tool permissions. Dangerous / not-pre-approved operations pop an Allow / Deny dialog — Deny is the default (Return denies) so an accidental keypress can't approve. Safe/allow-listed ops run without asking.
• esc. Interrupts the current turn; also cancels an in-progress connect.
• File ▸ New Conversation / Resume Last. Fresh conversation or resume the most recent one (when idle).
• ! bang commands. !somecommand runs a shell command directly on the proxy host (project dir), output streamed back — bypasses Claude, no permission dialog (you invoked it). Arbitrary shell exec, by design.
• View ▸ Dark Mode. White-on-black, session-only.
• Scrollback. Scroll up to read history; the view follows new output only when you're at the bottom.

Permissions model

The proxy runs the Agent SDK with your host's settings, honoring the allow-rules in ~/.claude and the project's .claude/settings*.json: safe/allow-listed operations auto-run; dangerous/unlisted ones prompt the Allow/Deny dialog on the SE.

---

Known limits

• Single client only (one SE at a time).
• Text is Mac Roman; characters with no Mac Roman equivalent render as ?.
• The proxy should not stream very large output while a permission ask is outstanding (the SE pauses reading during the modal dialog).
• Clipboard sharing is not yet implemented.

---

Project layout

src/                 SE app (C): event loop, UI, wire codec, transcript, netmac, proto
resources/MacCode.r  Rez resources: window, menus, dialogs, SIZE, app icon (ICN#/BNDL/…)
proxy/               Node/TypeScript proxy (Claude Agent SDK) + vitest tests
tests-native/        Host-side unit tests for the SE's pure-C logic (wire codec, transcript)
tools/               Build helpers (e.g. set_bundle_bit.py)
scripts/             setup / build / emulator-launch scripts
docs/                Toolchain + emulator setup guides
deps/                Retro68 toolchain + emulators (gitignored)

See the toolchain and emulator guides in docs/: RETRO68_SETUP.md, EMULATOR_SETUP.md, WORKFLOW.md.

---

Smoke test

connect → prompt → streamed reply (✻ verb, ● tool lines) → Allow a tool, Deny another → esc to interrupt → scroll back → File ▸ New / Resume → !ls → View ▸ Dark Mode → kill the proxy mid-session and reconnect via Session ▸ Connect → no crashes.

---

License

MIT — see LICENSE.