Claude Code Haha

// Overview

Claude Code Haha (cc-haha) is a desktop workspace for Claude Code that runs on macOS, Windows and Linux. Where the CLI gives you one terminal session at a time, cc-haha puts the same agent behind a graphical shell: multiple sessions side by side with global search across them, launching a task directly on a branch or a git worktree, a full-width syntax-highlighted diff review of what the agent changed, and a built-in browser preview so you can check an edited page without leaving the app.

Two things it adds are worth calling out. Permission approval is graphical rather than a terminal prompt, with the project and its permissions kept visible on the first screen, and model choice is per-session — the picker lists your own providers, saved presets and local endpoints together, covering Claude, ChatGPT, Grok and anything OpenAI-compatible you point it at. That makes it usable as a front end for a local model as readily as for a hosted one.

Around the core session view sit a set of management surfaces: visual managers for MCP servers and subagents, an Agent Teams workbench for multi-agent collaboration, dynamic workflow orchestration, model request tracing, Computer Use, image generation, and a skill marketplace that shows each skill's source and safety status before you install it. There are also conveniences aimed at long-running use — colour themes, desktop pets, scheduled tasks, IM integration, and H5 remote access for checking on a session from a phone.

The usual path is to download a signed installer from the Releases page and configure a provider, API key and default model on first launch; the documentation notes that unsigned Windows installers may trip SmartScreen and that draft macOS builds can need one-time manual approval. If you would rather debug the underlying CLI or server, the repository runs from source with Bun. The project is MIT-licensed and maintained by a single author in their spare time, with sponsorship from a couple of inference providers whose presets ship built in.

// What it does

• Multi-session workspace with global search across sessions
• Launch a task straight onto a branch or a git worktree
• Full-width syntax-highlighted diff review of the agent's changes
• Built-in browser preview for verifying an edited page in place
• Graphical permission approval instead of terminal prompts
• Per-session model choice across providers, saved presets and local endpoints — Claude, ChatGPT, Grok or an OpenAI-compatible URL
• Visual managers for MCP servers and subagents, plus an Agent Teams workbench and dynamic workflow orchestration
• Model request tracing, Computer Use and image generation in the same app
• Skill marketplace that surfaces each skill's source and safety status before installation
• Scheduled tasks, IM integration and H5 remote access for checking on a running session
• Ships for macOS, Windows and Linux; the CLI and server also run from source

// Getting started

Most people install the desktop build. Running from source is for debugging the underlying CLI or server.

https://github.com/NanmiCoder/cc-haha/releases

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

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Running several Claude Code sessions in parallel on different branches or worktrees without juggling terminal tabs
• Reviewing an agent's diff and previewing the edited page before accepting the change
• Approving tool permissions through a dialog rather than a terminal prompt, with the project's permission state visible
• Pointing a coding agent at a local or self-hosted model via a custom endpoint preset
• Managing MCP servers, subagents and skills visually instead of hand-editing configuration files
• Checking on a long-running task from a phone through H5 remote access

// How Claude Code Haha compares

Claude Code Haha alongside other open-source coding-agent harnesses & runtimes tools AI/TLDR tracks, ranked by GitHub stars.