cc-haha (Claude Code Haha) is a local-first, cross-platform desktop workspace for Claude Code and other coding agents, built for developers and teams who want multi-session agent work, Git worktree isolation, diff review, and multi-model routing inside one native application instead of a terminal.

What it is

Claude Code Haha is a desktop application for macOS, Windows, and Linux that wraps Claude Code and similar coding agents in a graphical workspace. It is written in TypeScript, built on Electron, licensed under MIT, and distributed as installers from GitHub Releases, with documentation hosted at cchaha.ai. It sits in the Anthropic Claude Code and MCP ecosystem: it manages MCP servers and SubAgents through visual managers, speaks the Model Context Protocol, and routes work to Claude, ChatGPT, Grok, presets, or local endpoints through a single provider list. Around that core it adds an Agent Teams workbench, dynamic Workflow orchestration, a skill marketplace, Computer Use, scheduled tasks, and model trace.

The concrete problem it solves is the terminal-only loop that coding agents normally require. Running an agent from a shell usually means several open terminals, no shared session state, permission prompts rendered as text, diffs read through git diff, and model switching done by editing configuration files. cc-haha replaces that workflow with a GUI: sessions launch against a branch or a Git Worktree so agent edits stay isolated, changes are reviewed in a full-width syntax-highlighted diff, tool-call permissions are approved in a dialog, and the edited page is verified in a built-in browser preview. The same application carries those sessions to a phone through H5 remote access and IM integration covering WeChat, Feishu, DingTalk, Telegram, and WhatsApp.

Key capabilities

• Multi-session workspaces with global search, keeping several agent sessions, projects, and permissions visible at once.
• Branch and Git Worktree launch, so an agent edits an isolated checkout rather than the main working tree.
• Diff review with full-width syntax-highlighted diffs, plus built-in browser preview of the edited page.
• GUI permission approval for tool calls, with stage-by-stage progress and model trace showing what ran.
• Model selection across Claude, ChatGPT, Grok, presets, and local endpoints, with image generation available.
• Visual MCP and SubAgent managers, an Agent Teams workbench, and dynamic Workflow orchestration.
• Computer Use, a skill marketplace, colour themes, desktop pets, and scheduled tasks in the same app.

Who uses it and how

• Individual developers on macOS, Windows, or Linux who run one agent per task in a separate Worktree and review the resulting diff before merging.
• Teams that need reviewable agent output: the diff view, stage-by-stage progress, and model trace make each run auditable without reading raw shell logs.
• Developers who want to follow a run away from the desk, using H5 remote access or an IM channel such as WeChat, Feishu, DingTalk, Telegram, or WhatsApp.
• Users who bring their own inference, pointing the provider list at Claude, ChatGPT, Grok, a preset, or a local endpoint and supplying an API key.
• Contributors debugging the underlying CLI or server, who run it from source rather than from an installer.

Getting started

Download the macOS, Windows, or Linux installer from Releases, then configure the model provider, API key, and default model in Settings on first launch. To run the CLI from source, use bun install, copy .env.example to .env, and execute ./bin/claude-haha.

How it compares

The facts provided name no paid products that cc-haha replaces and no directly comparable tools, so it stands alone in this registry. It is MIT licensed, self-hosted by nature as a desktop application, and keeps agent sessions local rather than routing the workspace through a vendor service.

When to use it — and when not to

Expect to supply your own model access: the app ships with settings for Claude, ChatGPT, Grok, presets, and local endpoints, and sponsors including Atlas Cloud and ApiSmart are available as providers, but the API key and billing relationship are yours. Installation carries friction worth knowing about, since public macOS releases require signing and notarization while unsigned Windows installers may trigger SmartScreen, in which case the documented path is "More info" then "Run anyway". It is a poor fit for anyone wanting a zero-configuration hosted service with no desktop application, and the project currently carries 189 open issues, so a self-hoster should expect to track an active backlog.

bun install
cp .env.example .env
./bin/claude-haha