cc-haha is a desktop workbench for Claude Code on macOS, Windows, and Linux, with a companion command-line client. It wraps agent sessions in a graphical app: multiple sessions with global search, sessions started on a chosen branch or an isolated Git worktree, per-file diff review of each turn's changes, an in-app browser preview, and graphical approval of permission requests.
Models are selectable per session. Claude, ChatGPT, and Grok accounts sign in directly, presets cover third-party APIs such as DeepSeek, Kimi, and Zhipu GLM, and local models connect through LM Studio or Ollama. The app also generates and edits images in chat.
Beyond single sessions, cc-haha manages MCP servers and subagents through a GUI, runs Agent Teams as a visual multi-agent workspace, and lets the model write and run orchestration scripts that schedule subagents in parallel or in sequence, with stage views and resumable runs. Computer Use lets an agent take screenshots, click, and type in other desktop apps, and on macOS it does so without taking over your mouse and keyboard.
Sessions are reachable away from the desk through a mobile H5 web view and through chat apps including Telegram, Feishu, WeChat, DingTalk, WhatsApp, WeCom, QQ, and Slack. Scheduled tasks, local token usage statistics, model request tracing, a skill marketplace, color themes, and optional desktop pets round out the app.
Features
- Multi-session workbench: tabs, project switching, terminal access, session history, and Cmd+K full-text search across sessions
- Branch and worktree sessions: each new session picks a repository branch and the current working tree or an isolated worktree
- Diff review: lists each turn's changed files with syntax-highlighted diffs, and the whole turn can be undone
- Permission modes: five levels from asking for every permission to skipping them, with approvals handled in the desktop app
- Model choice: Claude, ChatGPT, and Grok sign-in, presets for DeepSeek, Kimi, and Zhipu GLM, and local models through LM Studio or Ollama
- MCP management: add, edit, and remove MCP servers over STDIO, Streamable HTTP, or SSE at project, shared, or global scope
- Subagents and Agent Teams: create and tune subagents in a GUI and watch multi-agent teams, tasks, and messages in swimlanes
- Dynamic workflows: the model writes and runs orchestration scripts that schedule subagents in parallel or in sequence, with interrupt and resume
- Computer Use: agents take screenshots, click, and type in desktop apps, without taking over the real mouse and keyboard on macOS
- Skill marketplace: discover, preview, and install third-party skills from ClawHub and SkillHub, with source and safety status shown
- Remote access: a mobile H5 view plus Telegram, Feishu, WeChat, DingTalk, WhatsApp, WeCom, QQ, and Slack for chatting and approving permissions
- Scheduling and tracing: scheduled tasks in separate sessions, local token usage trends, and searchable model request logs
