Website: claudedeck.org

A self-hosted web application for visualizing and managing local AI coding agents. Provides a unified interface for Claude Code configuration, Codex CLI configuration, MCP servers, plugins, slash commands, hooks, agents, permissions, usage tracking, session transcripts, Agent Bridge, and other local agent extensions.

Claude Code starts simple, then slowly sprawls across config files and directories: ~/.claude.json, ~/.claude/settings.json, .mcp.json, slash commands, agents, skills, project settings, transcripts, and usage data. That works fine at small scale, but once your setup gets serious it becomes hard to see the whole picture, change things confidently, or understand what is actually configured.

Claude Deck gives you one local interface for that sprawl. It also has provider-aware Codex CLI support for tmux sessions, safe TOML configuration, feature flags, diagnostics, MCP/plugin inventory and supported CLI-backed mutations, and redacted export-only backups.

Claude Deck is best for people running multiple Claude Code or Codex CLI sessions, MCP servers, custom commands, hooks, agents, or tracking Claude Code usage across sessions.

If you only use Claude Code casually with mostly default config, Claude Deck may be overkill.

• Local only — no cloud
• No account — nothing to sign up for
• No telemetry — no usage tracking sent anywhere
• Works with your real files — reads and writes existing Claude Code and Codex config files

• Dashboard — Overview of local agent configuration with Claude Code context window visualizer
• Provider Switcher — Move between Claude Code and Codex CLI surfaces without leaving the app
• Config Editor — Browse, inspect, and edit Claude Code JSON settings or Codex TOML settings, including Codex profiles, runtime options, and feature flags
• MCP Servers — Add, edit, test, and manage MCP server connections with OAuth support. Browse and install servers from the MCP Registry. View tools, resources, and prompts. Supports stdio, HTTP, and SSE transports
• Slash Commands — Browse, create, and edit custom commands (user and project scope)
• Plugins — Browse installed plugins with detail views and enable/disable toggles; Codex plugins support CLI-backed inventory, install, and remove where the installed Codex CLI exposes safe commands
• Hooks — Configure automation hooks by event type (PreToolUse, PostToolUse, etc.)
• Permissions — Visual allow/deny rule builder for tool access control
• Agents — Create and manage custom agent configurations
• Skills — Browse installed skills and discover new ones from skills.sh
• Memory — View and edit Claude Code memory files
• Output Styles — Configure response output formats
• Status Line — Customize Claude Code status line display
• Agent Bridge — Discover and monitor Claude Code and Codex CLI sessions running in tmux. Attach up to 4 terminals simultaneously in a 2x2 grid with independent read-only/interactive modes, fullscreen toggle, and per-pane controls. Spawn new sessions and manage provider-specific options directly from the UI
• Agent Mail — Coordinate local Claude Code and Codex CLI agents through durable per-repo identities, structured context requests, handoffs, and an inspectable team mailbox
• Session Transcripts — View conversation history with full message details and tool use
• Usage Tracking — Monitor token usage, costs, and billing blocks with daily/monthly charts
• Plan History — Browse and review Claude Code implementation plans
• Backup & Restore — Create and manage Claude Code backups with selective restore, plus redacted export-only Codex backups
• Projects — Discover and manage project directories

Codex CLI support has moved from experimental plumbing to a usable provider surface:

• Provider-aware Agent Bridge can discover, spawn, resume, fork, attach to, and kill Codex tmux sessions.
• The Codex config editor now handles safe TOML settings, profiles, runtime controls, and feature flags from codex features list.
• Codex settings include dropdowns for known enum values and help tooltips for settings and feature flags where official descriptions are available.
• Codex MCP and plugin inventory are visible, with supported CLI-backed add/remove or install/remove actions.
• Codex exports are redacted and export-only by design.
• Project discovery is now easier from the UI, including directory browsing when adding projects.

Codex support is explicit about provider boundaries: usage/context parity and session transcript browsing are not supported for Codex yet; history and model-cache diagnostics avoid prompt text and raw cache payloads; Codex automatic restore is refused because exports intentionally exclude auth, history, cache, and local state.

Agent Bridge	Dashboard
Monitor and interact with Claude Code and Codex tmux sessions	High-level overview of your local agent workspace

Config	MCP Servers
Edit safe Codex TOML settings and inspect provider diagnostics	Manage MCP connections, status, and configuration

Usage Tracking	Session Transcripts
Cost visibility, charts, and billing blocks	Browse conversation history and tool usage details

Skills
Browse installed skills and discover new ones

git clone https://github.com/adrirubio/claude-deck.git
cd claude-deck
docker compose up

This builds and starts Claude Deck at http://localhost:8000, mounting your ~/.claude directory and ~/.claude.json configuration file. Codex support reads $CODEX_HOME, defaulting to ~/.codex, when available in the runtime environment.

Prerequisites: Python 3.11+, Node.js 18+

git clone https://github.com/adrirubio/claude-deck.git
cd claude-deck
./scripts/install.sh

./scripts/dev.sh

This starts:

• Backend at http://localhost:8000 (API docs at http://localhost:8000/docs)
• Frontend at http://localhost:5173

To stop or restart the dev servers for this checkout:

./scripts/dev.sh stop
./scripts/dev.sh restart

To make the dev environment reachable from another machine on your LAN or tailnet (e.g. to monitor tmux sessions via Agent Bridge from a different host), pass --host:

./scripts/dev.sh --host 0.0.0.0

Both servers will then bind to all interfaces.

When running Claude Deck on several machines, set a display name and accent color so each browser window clearly identifies the backend it controls:

CLAUDE_DECK_INSTANCE_NAME="Studio Mac" \
CLAUDE_DECK_INSTANCE_ACCENT="blue" \
./scripts/dev.sh --host 0.0.0.0

Supported accents are blue, green, purple, orange, red, pink, cyan, and slate. The name appears in the header, browser tab title, Agent Bridge terminal panes, and destructive confirmations.

To preview the documentation site:

./scripts/docs-dev.sh

This starts VitePress at http://localhost:5174/docs/. Use --host 0.0.0.0 if you need to reach it from another machine.

For a release check, ./scripts/build.sh builds both the app frontend and the documentation site.

Claude Deck reads and writes these Claude Code configuration files:

Codex CLI support uses $CODEX_HOME, defaulting to ~/.codex:

See CONTRIBUTING.md for setup, style, and PR guidelines.

API documentation is available at http://localhost:8000/docs when running the dev server.

If you use Claude Code heavily, issues and feature requests are especially welcome.

Adrian (13) and Juan during the 2025 Christmas break as a learning project — to explore open source, Claude Code, and full-stack development together.

The session transcript viewer was inspired by and includes code adapted from claude-code-transcripts by Simon Willison.

The usage tracking feature ports algorithms from ccusage by ryoppippi, including session block identification, tiered pricing, and burn rate projections.

Claude Deck is a community project and is not affiliated with or endorsed by Anthropic.

MIT License