Route Claude Code through your Mac's Neural Engine. Free, private, and fast enough for everyday refactors and code review. No cloud round-trip, no per-token billing; Apple Intelligence runs entirely on-device.
Launch ToolPiper. Open the menu-bar icon → Endpoints → "Add Endpoint" → pick Apple Intelligence. The Neural Engine backend lights up immediately, no model download, no warm-up. Name it whatever you want; the recipe assumes "Apple Intelligence (local)".
Ask ToolPiper in chat to "set up Claude Code", or call the claude_code_install MCP tool directly. ToolPiper symlinks the bundled claude-tp helper onto your PATH (~/.local/bin/claude-tp), mints a scoped inference token, and writes the helper's own mode-0600 config file. It never touches ~/.claude/settings.json. The install is idempotent, so running it again just refreshes the symlink and token.
Open a new terminal window (a shell that was already open before the install won't see the updated PATH). The claude-tp helper carries the config: it reads ToolPiper's live endpoint list, builds a per-launch settings file, and hands off to Claude Code.
claude-tp
# In Claude Code, type "/model" to confirm Apple Intelligence appears.Claude Code's /model picker now lists your ToolPiper endpoints. Select Apple Intelligence; every subsequent prompt routes through the Neural Engine. Confirm with "What model are you?".
Type "switch to my local Qwen one" or "use my OpenAI endpoint"; ToolPiper's MCP endpoint_set tool reroutes the current shell. No file editing, no env var dance.
Apple Intelligence runs on the Neural Engine. Your prompts and code never leave the machine. Same hardware Apple uses for on-device summarization.
No model download, no GPU warm-up. ToolPiper exposes Apple Intelligence the moment macOS does; first prompt is instant.
There's no token meter. Run Claude Code as much as you want; Apple Intelligence is included with macOS.
Apple Intelligence is built for fast on-device work: summaries, refactors, code review, mechanical edits. For multi-step reasoning or large-context refactors, switch to a stronger backend mid-session via <code class="font-mono text-xs text-fg-muted">"switch to my OpenAI endpoint"</code>. ToolPiper makes that one sentence; you keep your Claude Code session.
Yes. ToolPiper's Anthropic proxy translates Apple Intelligence's output into the Anthropic content-block shape Claude Code expects, including <code class="font-mono text-xs text-fg-muted">tool_use</code> blocks. Bash, Edit, Read, Write all flow through.
No. The claude-tp helper carries a ToolPiper-minted token in its own config file, not an Anthropic key. The token validates locally against ToolPiper's token store; it never reaches Anthropic.
Ask ToolPiper to uninstall it, or call the <code class="font-mono text-xs text-fg-muted">claude_code_uninstall</code> MCP tool. It removes the <code class="font-mono text-xs text-fg-muted">~/.local/bin/claude-tp</code> symlink, deletes the helper's config file, and revokes the bearer token. Your <code class="font-mono text-xs text-fg-muted">~/.claude/settings.json</code> was never touched, so there's nothing to restore there.
ToolPiper is a free download. Configure once and Claude Code routes through your Mac.