claude-config-ui

Claude Config Manager

The mission control for Claude Code.

Run a terminal grid of live Claude agents, edit every config file from one place, watch your usage limits tick down, move your whole setup between machines encrypted — all in one fast, local, open-source desktop app. Nothing leaves your machine.

   

or, in any terminal:  npx claude-config-ui

▶ Watch the 40-second tour

Files saved by this tool land in Claude Code's standard locations — Claude Code picks them up automatically on the next session start. No daemons, no symlinks, no special integration step.

⌘ The Claude Code workspace

The headline feature: a full terminal environment for Claude Code, inside the app. Stop juggling terminal tabs — run your agents in a grid and watch them all at once.

• Real terminals, embedded — spin up as many shells (or one-click Claude sessions that launch the claude CLI) as you want. Actual PTYs, full color, resize, scrollback.
• Split 1 / 2 / 3 / 4 ways — a layout switcher tiles panes so you can run four agents side-by-side and multitask like a cockpit.
• A file tree that follows your work — a Nimbalyst-style workspace tree on the left. Click a pane (or cd into a project) and the tree jumps to that folder automatically so its files are right there. "Open terminal here" on any folder.
• A control bar for Claude — a bottom strip that drives the focused session the way you would: pick the model and effort, cycle permission mode (⇧⇥), toggle extended thinking (⌥T), and fire /context, /compact, /clear, /help — no typing.
• Drag & drop paths → drag any file/folder from the workspace tree (or from Finder) straight onto a terminal pane and its absolute path pastes in — no copying long paths by hand.
• Transcripts that survive restarts — every session is saved to disk. A History browser lets you re-open any past session's transcript (colors preserved) or reopen its folder in a fresh terminal. Quit the app mid-run? Your record is still there.
• A built-in file viewer — click any file in the tree to open it inline:
  • Markdown — rendered preview ⇄ editor, save with ⌘S
  • HTML — live sandboxed preview ⇄ source
  • PDF — native viewer
  • Images & code — inline
• Drop-to-read, anywhere — drag a .md / .pdf / text file from Finder onto any view (or from the workspace tree onto the header) and it opens in a full-screen reader. Read-only, instant, Esc to close.

📊 Live usage meter

A gauge next to the header shows exactly how much you have left — pulled from Claude's own local usage cache, so it matches the app precisely:

• 5-hour window, weekly, and per-model weekly (Fable 5 / Opus) remaining
• Live "resets in…" countdowns for each
• Color-coded: comfortable → low → critical
• Home shows the full picture — a token strip on the dashboard with your all-time tokens used (computed locally from your session transcripts), plus 5-hour, weekly, and Fable 5 weekly remaining at a glance

No more guessing when you'll hit a limit mid-flow.

Everything else it does

⚙️ Config, all of it	Edit settings.json, CLAUDE.md, .mcp.json, subagents, slash commands, output styles, keybindings — across all four scopes, every field with a tooltip that says what it does and why it matters.
🔑 Credentials once	Set GitHub / AWS / Vercel / Stripe / 28+ services one time at user scope; they're available to every project. Masked inputs, deep links to each token page.
🧭 Agent Orchestrator	A live board to launch one agent or a whole team, watch the tool/sub-agent tree render in real time, and resume any run. Discovers agents already running on your machine.
📦 Encrypted Transfer	Pack your entire Claude world into one AES-256-GCM .ccsync file and restore it on another laptop with a per-file preview.
🗂️ Projects & sessions	A bird's-eye view of every project that uses Claude, plus a per-project decision log (logic.md) and key tracker (credentials.md).
💬 Conversation archive	Saves a verbatim Markdown copy of every Claude Code conversation, grouped per project — a readable local record that costs zero context-window tokens.
🏠 Home dashboard	Opens on a control center: KPI tiles + Recommended next steps derived from your actual setup (lock down credentials, enable the sandbox, review a risky bypassPermissions default…).

Install

Mac app

Download from the latest release:

• Claude Config-X.Y.Z-arm64.dmg — Apple Silicon (M1 / M2 / M3 / M4, 2020+)
• Claude Config-X.Y.Z-x64.dmg — Intel Macs (2019 and earlier)

Open the .dmg, drag Claude Config to Applications, launch from Spotlight or Launchpad.

First launch (one time): the app is signed ad-hoc but not notarized (no paid Apple Developer certificate), so macOS Gatekeeper blocks it once. Right-click the app → Open, then Open in the dialog. Or clear the quarantine flag:

xattr -dr com.apple.quarantine "/Applications/Claude Config.app"

The "app is damaged" seal bug from builds ≤ v0.9.0 is fixed in current releases.

Windows app

Download Claude-Config-X.Y.Z-x64-setup.exe from the latest release, run the installer, launch Claude Config from the Start menu.

First launch (one time): Windows SmartScreen blocks the unsigned installer. Click More info → Run anyway. Standard for indie/OSS apps without a code-signing cert.

Command-line (any OS)

npx claude-config-ui

The launcher picks a free port, opens your browser, and detects your OS-correct paths automatically. Ctrl+C to stop.

The embedded-terminal, drag-and-drop, and macOS onboarding features are richest in the desktop app (they need OS access the browser sandbox doesn't allow). Everything else runs identically over npx.

Why it exists

Claude Code reads config from 30+ possible files across user, project, project-local, and enterprise scopes. Most people learn the system by trial-and-error or by copy-pasting community gists. This turns that into a guided, visual workflow — and then goes further, giving you a place to actually run and watch your agents.

• Auto-detected paths — ~/.claude/ on macOS/Linux, %APPDATA%\Claude\ on Windows, project paths from your repo, enterprise paths from /Library/Application Support/ClaudeCode/ or /etc/claude-code/.
• Community presets baked in — Karpathy's 4-rules CLAUDE.md, HumanLayer's skeleton, Trail of Bits credential lockdown, popular hook recipes — one click.
• Auto-backup — every save writes a *.bak-<timestamp>. Nothing is lost.
• Path sandbox — the API refuses to read or write anything outside your home directory, cwd, or known Claude Code enterprise dirs.

The four scopes

Mirrors how Claude Code actually resolves config:

Tab	What	Path
Global Claude	Your personal user-level config (all projects)	~/.claude/
Global Project	Team-shared project config (committed to git)	<project>/.claude/ + CLAUDE.md + .mcp.json
Local Claude	Personal overrides for one project (gitignored)	<project>/.claude/settings.local.json + CLAUDE.local.md
Project Local (Enterprise)	Organization policy deployed by IT (highest precedence)	/Library/Application Support/ClaudeCode/ · C:\Program Files\ClaudeCode\ · /etc/claude-code/

Precedence (later wins): user → project shared → project local → enterprise.

What it edits

• settings.json — every key in the Claude Code 2.1.28x settings reference, in 16 collapsible, filterable groups: model & reasoning (effort caps, fallback chain, fast mode, ultracode, prompt-cache lifetime), permissions and auto-mode classifier rules, the full sandbox (filesystem, network, credentials), MCP allow/deny lists, env, memory & context, interface & terminal, status line, git attribution, plugins & skills, agents & worktrees, Remote Control & notifications, auth helpers, updates & privacy, and managed-only policy keys. Toggles show Claude Code's real default when a key is unset.
• Hooks builder — all 33 lifecycle events (incl. PreModelSwitch / PostModelSwitch), matcher hints per event, and all five handler types: command, http, mcp_tool, prompt, agent.
• Credentials (Global Claude only) — AWS, GCP, Azure, Cloudflare, Fly, Vercel, Netlify, GitHub, GitLab, Postgres, Supabase, Neon, MongoDB, Redis, Slack, Linear, Notion, OpenAI, Anthropic, Stripe, Sentry, Datadog and more — masked inputs, eye-toggles, deep links, written into the env block of ~/.claude/settings.json.
• CLAUDE.md / CLAUDE.local.md — markdown editor with line/char counts, 12 section templates, and four full-doc presets (Karpathy, HumanLayer, Trail of Bits, Minimal).
• .mcp.json (stdio / http / sse / ws, OAuth, headers helper, per-server timeout) · keybindings.json (all 24 contexts, null to unbind) · agents/ · commands/ · skills/ · output-styles/ · managed-settings.json — all guided, all templated.

Agent Orchestrator

A live, visual team board for running Claude agents. Launch one on a task, or launch several at once — each runs concurrently as its own card.

• Crews — CrewAI-style pipelines, zero code — build reusable multi-agent pipelines entirely in the UI: define agents (role / goal / backstory / model / permissions), give them tasks in sequence, wire which task's output feeds which with checkboxes, and watch a live flow diagram redraw as you edit. Save the crew, then Run it any time — it asks only for your declared {inputs} and executes step by step, handing each agent's output to the next, with per-step status, outputs, and cost. Sequential or manager-delegated (hierarchical) modes.
• See what's already running — discovers claude sessions you didn't launch here (other terminals/windows) and shows each one's activity, model, project, and a claude --resume command.
• Run a team preset — pick the general agent or any subagent definition; choose model, permission mode, working dir, turn cap. Run a team template (Build squad / Ship crew / Research pod / Bug hunt) orchestrated (a lead delegates) or parallel.
• Campaigns — multi-week work with a self-updating plan that survives restarts; hit Run next session to pick up exactly where it left off.
• Watch the hierarchy — every tool call, skill, MCP call, and spawned sub-agent renders as a color-coded tree as it happens. Live cost, tokens, turns per card; a global skill-activity feed.
• Measurement — aggregate finished runs: success rate, cost, tokens, breakdowns by agent and model.

Under the hood it spawns headless claude -p … --output-format stream-json and parses the event stream — no extra dependencies, using your already-authenticated claude CLI.

MCP servers

A dedicated MCP tab across all three scopes. See every server with its transport/auth/alwaysLoad state, enable/disable to save context (reversibly stashed, with a backup), add/edit/remove with guided fields, and resolve issues with per-server diagnostics that give the exact copy-paste fix (/mcp, claude mcp get, reset-project-choices). Secrets masked.

Transfer — move your setup to another machine

Pack your whole Claude world into one encrypted .ccsync file and restore it elsewhere.

• What goes in — live-counted categories: ~/.claude.json (MCP servers + API keys), settings, global CLAUDE.md, agents, skills, commands, keybindings, per-project auto-memory, and the app's own data. Add extra files/folders — .env files ride along.
• Real encryption — sealed with AES-256-GCM, key derived from your passphrase with scrypt. Safe over AirDrop, iCloud Drive, or a USB stick.
• Safe restore — a per-file preview (new / changed / identical); anything you overwrite is backed up first. Restores refuse paths outside your home directory.
• Login stays out on purpose — the OAuth token is per-machine; run claude + /login once on the new laptop.

Projects, sessions & memory

A bird's-eye view of every project that uses Claude — discovered by scanning your code roots and decoding the working directories of recent Claude Code sessions. Edit each project's local files inline (CLAUDE.md, .claude/settings.json, .mcp.json, …).

• Claude sessions — the same list /resume shows, with first prompt, last-worked time, message count, size, and detected project. Sort, rename, reassign, or delete throwaway sessions to reclaim disk.
• logic.md — a per-project decision log in a central context vault, so rationale isn't re-litigated across sessions. Toggle Auto-maintain to have Claude read + append it every session.
• credentials.md — a local-only, masked inventory of the keys each project uses, so you know what exists and when to rotate.
• Conversation archive — one click (or auto, on app open) writes a verbatim Markdown copy of every Claude Code conversation to ~/.claude-config-ui/conversations/<project>/, in plain "You / Claude" turns with no summarization. It only reads the session files Claude already wrote and writes separate .mds — it never feeds anything back to Claude, so it uses zero context-window tokens.

Design

A redesigned, high-contrast mechanical interface: a monochrome canvas with a disciplined primary-color accent system — blue for interactive/active, red for live/destructive, yellow for warnings — over a precise engineering-grid backdrop. Animated view transitions, count-up metrics, and skeleton loaders throughout, all honoring prefers-reduced-motion. A motivational line greets you on open, and a first-run macOS gate walks you through the folder permissions Claude needs to create projects.

Requirements

• Node.js 20.9 or newer
• macOS, Windows, or Linux

Alternative install paths

From source:

git clone https://github.com/HarshitSingh-PM/claude-config-manager.git
cd claude-config-manager
npm install
npm run dev        # → http://localhost:3000

Global install:

npm install -g claude-config-ui
claude-config-ui   # or: ccm

The npm package is claude-config-ui (the bare name was taken); the GitHub repo keeps claude-config-manager.

Configuration

Everything is auto-detected. The only thing you set is the project directory for project-scoped tabs (defaults to where you launched). No .env, no config file.

• PORT — override the default 3737.
• CCM_NO_OPEN=1 — don't auto-open the browser.

Safety

• Every write makes a timestamped .bak-* first (autosave takes one per file per session).
• The file API refuses paths outside ~, cwd, and known enterprise dirs; the same sandbox applies to the terminal file tree and viewer.
• Terminal transcripts are stored locally under ~/.claude-config-ui/terminals/ (most-recent 100 retained).
• Credentials are written to the env block of ~/.claude/settings.json — plain text on disk. Fine for personal API keys; for high-blast-radius secrets use Claude Code's apiKeyHelper instead.

Stack

• Next.js 16 (App Router, Turbopack, React 19.2) — standalone server output for npx
• TypeScript strict · Tailwind CSS 4 · Framer Motion · Radix UI · Lucide
• node-pty + @xterm/xterm for the embedded terminals (N-API prebuilds — no native rebuild)
• Electron for the desktop .dmg / .exe distribution (the npm package is pure Node)

Building desktop apps yourself

npm install
npm run dmg          # macOS arm64 .dmg → dist-electron/
npm run exe          # Windows x64 NSIS .exe → dist-electron/

Cross-compiling to Windows from Apple Silicon needs Rosetta (sudo softwareupdate --install-rosetta --agree-to-license). Builds are unsigned by default — set mac.identity / win.certificateFile in package.json#build to sign for distribution.

Contributing

See CONTRIBUTING.md. Easy wins: add a service to the Credentials catalog, a CLAUDE.md section template, or a subagent / slash-command template (src/lib/presets/).

License

MIT. See LICENSE.

Acknowledgements

Schema and presets drawn from the official Claude Code docs, awesome-claude-code, trailofbits/claude-code-config, awesome-claude-code-subagents, wshobson/commands, HumanLayer's "Writing a good CLAUDE.md", Karpathy's CLAUDE.md gist, and claude-code-hooks-mastery. Terminal file-tree UX inspired by Nimbalyst.