A macOS desktop app that lets you watch what Claude Code is doing in your terminal as a top-down office scene. It treats your local ~/.claude/ directory as the single source of truth, drawing each agent as an “employee” and each running session as the “orchestrator,” animating what is happening in real time. It is built by t-soda and was first released on June 27, 2026 as an open-source project under the MIT license. The app is built with Tauri v2 (Rust) plus React and Pixi.js, and makes no external network calls at all.
Claude Code’s Hooks, Skills, and Agents are spread across several config locations, which makes it hard to answer “what is this session actually composed of right now?” Claude Code Park aims to show that answer at a glance.
Key Features
- Per-session hook visualization: See the hook configuration that actually applies to a given session — including hooks provided by plugins — collected in one view, without jumping between config files
- Watch hooks fire as animations: When and which hook ran is drawn on screen, so you can grasp the hook lifecycle just by watching real behavior. It also answers the question “is the hook I wrote actually firing?”
- Skills and Agents at a glance: See which agents and skills are defined, and which of them are currently “at work”
- Manage multiple sessions on one screen: Track the progress of several Claude Code sessions running in parallel from a single view
- One click back to the terminal: Click an employee (session) to bring the terminal that launched it to the front. Terminal.app, iTerm2, VS Code, and Ghostty are supported
- Management dashboard: Utilization, call count, failure rate, and token usage per agent, aggregated over today / 7 days / 30 days
The safety design is stated explicitly. It reads only your local ~/.claude/ — no telemetry, no accounts, no network calls. It keeps no separate database and simply reflects your config files live, so there is no hidden state. Config writes are backed up first and then applied through a temp file plus an atomic rename, and when editing settings.json it replaces only the hooks key, leaving permissions, env, and everything else untouched.
Pricing
| Plan | Price | Highlights |
|---|---|---|
| Open source | $0 | All features available. No account and no subscription (MIT license) |
Pricing is as of August 2026. Please check the official repository for the latest information.
Distribution is a .dmg download from the GitHub Releases page. The app is unsigned, so on first launch you need to clear the macOS Gatekeeper warning manually (right-click and choose “Open,” or allow it from System Settings → Privacy & Security). Building from source requires Node.js 18 or later and Rust (stable).
Pros & Cons
✅ Pros
- Scattered Hooks, Skills, and Agents settings can be reviewed per session in one place
- You can visually confirm whether a hook really fired, which speeds up debugging your setup
- Even when running several sessions at once, you can see which ones are waiting
- Zero external communication and fully local operation, making it easy to adopt in environments that handle work code
- Free and MIT-licensed, so the behavior can be verified in the source
⚠️ Cons
- macOS only. Because terminal focusing relies on AppleScript and
ps, it does not run on Windows or Linux - Requires Claude Code to be installed with an existing
~/.claude/. It does not support other AI coding tools - The app is unsigned, so the first launch requires clearing Gatekeeper manually
- It is an individual open-source project, so update cadence and support cannot be assumed to match a commercial product
Comparison with Similar Services
| Item | Claude Code Park | Claudia | ccusage | Claude Code itself |
|---|---|---|---|---|
| Form factor | Desktop app | Desktop app | CLI tool | CLI |
| Main purpose | Visualizing sessions and Hooks/Skills/Agents | Session and agent management plus usage analytics | Aggregating token usage and cost | Doing the development work itself |
| Supported OS | macOS only | Windows / macOS / Linux | Anywhere Node.js runs | macOS / Linux / Windows |
| Hook firing visualization | Yes | ─ | ─ | Check via logs and output |
| License | MIT | AGPL-3.0 | Open source | Anthropic product |
| Price | Free | Free | Free | Included in Claude plans |
Claudia is a GUI that broadly covers session management and usage analytics, and ccusage is a CLI focused on cost aggregation. Claude Code Park differs in that its center of gravity is “following the composition of a session and the behavior of its hooks with your eyes.”
Who Is It For
- People writing hooks to enforce team conventions who are unsure whether those settings are really in effect
- People running several Claude Code sessions in parallel and losing track of which one has stalled
- People who want to see per-agent utilization and token usage in order to revisit how they delegate work
- People who are just starting to learn how Hooks, Skills, and Agents work and want to understand them by watching real behavior
- Mac users who want a fully local tool that sends nothing outside their machine
Summary
Claude Code Park translates the inner workings of Claude Code — which you could previously only follow through text logs — into an easy-to-read top-down office. The hook configuration and firing visualization in particular will reliably cut down the time you spend staring at config files. The macOS-only requirement and the dependency on Claude Code are real constraints, but since it is free and runs entirely locally, it is worth installing once you start using Hooks and Agents in earnest.