CLAUDE CODE WEB UI · SELF-HOSTED · 2026
Claude Code Web UI — Dashboards, Remote Control, and the Canvas
"Claude Code web UI" means three different things in 2026: Anthropic's official claude.ai/code surfaces (web + mobile, driven by Remote Control), the self-hosted open-source dashboard wave — Claude Deck, Clauster, claude-pulse, claudex, yepanywhere — that turns your browser or phone into a cockpit for local agent sessions, and VibeFuse, which skips the window-into-a-terminal metaphor entirely: agents live as widgets on a desktop canvas you arrange yourself.
Layer 1 — the official web UI: Remote Control
Anthropic's own answer shipped February 25, 2026 as a research preview: run
claude remote-control (or /remote-control
inside a session), and claude.ai/code or the Claude mobile app becomes a window into that
session. Execution and filesystem access stay on your machine — the local process makes
outbound HTTPS only and never opens inbound ports — while the session transcript syncs
through Anthropic servers so you can keep steering from your phone. It is available on
Pro, Max, Team, and Enterprise with a claude.ai login; API keys do not work,
and one flag catches exactly the developers most likely to want it: setting
DO_NOT_TRACK or DISABLE_TELEMETRY
silently disables the feature-flag evaluation Remote Control depends on. Know its limits
before building a workflow on it: it drives one session, it ends if the local
process closes, and a machine that loses network for roughly ten minutes times the
session out entirely. There is no dashboard view of many projects, no config editor, and
no approval queue across agents — that is the gap the self-hosted wave fills.
Layer 2 — the self-hosted Claude Code dashboard wave
2026 produced a dozen open-source projects that give Claude Code (and usually Codex) a browser interface on your own hardware. They split into two families: cockpits, which spawn and drive real CLI sessions from a phone or browser, and control panels, which visualize and manage configuration, transcripts, and usage. All verified September 2026.
| Project | What it is | Standout |
|---|---|---|
| Claude Deck (claudedeck.org, Apr 2026) | Local-only web dashboard for Claude Code + Codex CLI: MCP servers, slash commands, hooks, permissions, skills, usage tracking, session transcripts, backups | No account, no telemetry; Agent Mail + Agent Teams for multi-agent coordination; reads and writes your real config files |
| Clauster (PyPI clauster 0.2.1) | Self-hosted dashboard that spawns claude remote-control bridges on a homelab, NAS, or VPS — one card per project, attach from claude.ai/code or your phone by QR | Session survives laptop closing (runs on the host); loopback by default, password/reverse-proxy auth, no telemetry |
| claude-pulse (247★, MIT) | Zero-dependency local dashboard watching every Claude Code and Codex session via the .jsonl files they already write | Approve tool calls from your phone with Allow/Deny buttons — works over cellular, no Wi-Fi setup |
| claude-code-dashboard (kevin101681, Jul 2026) | Windows-first control room: persistent ConPTY sessions owned by server.js, drive claude.exe and codex from any browser | Sessions survive tab closes and PC restarts (watchdog + claude --resume); ntfy push when approval is needed; PWA |
| claudex (May 2026, MIT) | Mobile-first cockpit that spawns the real claude CLI as a subprocess — your keys, your config, your diffs | TOTP two-factor with one-shot recovery codes; plan tracking and parallel subagent strips |
| yepanywhere (MIT) | Self-hosted web UI for Claude and Codex with end-to-end encryption, push notifications, file uploads, no accounts or database | Server-owned processes — disconnecting does not interrupt the agent; optional public relay for remote access |
| CloudCLI (claudecodeui) (AGPL-3.0) | Web and mobile UI wrapping Claude Code, Cursor CLI, Codex, and OpenCode; self-hostable via npx or Docker, €7/mo managed cloud | Broadest CLI coverage; independent review flags thin auth middleware and the AGPL obligation for internal deployments |
| claude-code-webui (sugyan, MIT) | Streaming chat web interface over the claude CLI — the original npm-install web UI | Permission-mode switching, project picker, binaries for every platform; development slowed since late 2025 |
Project details as documented on each project's README/site (September 2026). This wave moves fast — verify star counts, auth posture, and maintenance status before committing a workflow.
Be honest: when a self-hosted dashboard is the right tool
Credit where due — if you already run Claude Code in a terminal and just need to reach it from elsewhere, the dashboards are excellent, free, and inspectable. claude-pulse is 247 stars of zero-dependency Node that does exactly one thing (approve tool calls from your phone) and does it well. Clauster is the right answer if your agents live on a homelab box and you want them attachable without SSH. Claude Deck is the best configuration explorer for the MCP/hooks/skills sprawl that Claude Code accumulates. None of these cost anything, all run on hardware you own, and the best of them are genuinely careful about it (Claude Deck ships a loud trust-model warning because it edits your real files). If your setup is "one terminal, many sessions, approve from the couch," ship a dashboard and stop reading.
What they share: the window metaphor
Every project in the wave — and Anthropic's Remote Control — is a window or cockpit over a terminal session running somewhere. The agent UI is still a chat transcript scrolling in a pane. When you run three agents on three worktrees, you are tab-switching between three windows.
The security tax
A web UI over a shell is a shell on a port. claude-code-web's README warns "no authentication mechanism — anyone with access to the port can control your computer"; independent review of CloudCLI flags a single-file auth middleware. Loopback-only binds and reverse proxies with auth are the standard mitigations — configure them or expose nothing.
Where the canvas differs
VibeFuse is not a window into a terminal — it is the desktop the agents live in: Claude Code, Codex, Gemini, Cursor Agent, and Qwen as live PTY widgets beside browser, file, video, and notes widgets. The layout is the interface, and Vibe Agent + Jarvis rearrange it by command.
Layer 3 — VibeFuse: the widget canvas a web UI can't be
A dashboard asks "what is my agent doing?" A canvas asks "what should this desk look like?" VibeFuse is a free Windows 10/11 harness where every agent is a widget you place: live terminals, browser tabs, file trees, video, notes, and MCP tools attached at the canvas level so every agent reaches them. The agent UI widgets layer 2026 standardized — tool-call cards, plan timelines, approval prompts — is not something VibeFuse renders inside a chat; it is the native form of the whole desktop. It is the first ever free widget-based AI harness, and it ships with an open-source marketplace where users make money selling widgets, skills, and styling packs with 80% payouts — the dashboards have no equivalent surface, and neither does Anthropic's claude.ai/code.
Practical differences that matter daily: everything runs local and offline — no relay, no transcript sync through anyone's servers, nothing to expose on your network because it is a desktop app, not a web server. Multi-agent is native: five CLIs side by side on named resumable sessions instead of one window per agent. Voice is local — Whisper + Piper on-device — so you dictate into whichever widget has focus. And because your Claude subscription still drives the CLI, pairing VibeFuse with Remote Control for phone-steering is a legitimate hybrid: canvas at the desk, window in your pocket.
Choosing between the agent CLIs themselves is covered in our OpenCode vs Claude Code and Claude Code vs Cursor comparisons; if you want Claude Code with a graphical wrapper specifically (not a dashboard), see Claude Code GUI. The config layer every one of these tools reads is documented in our AGENTS.md template.
Run the canvas instead of a window
A free widget-based AI harness for Claude Code, Codex, Gemini, Cursor Agent, and Qwen on Windows — local/offline, MCP at canvas level, and an 80%-payout creator marketplace.
Also compare the best vibe coding tools, the Codex CLI vs Claude Code head-to-head, and our Xirp alternative for Windows.
- ✓ Claude Code + Codex + Gemini + Cursor + Qwen widgets
- ✓ Named resumable sessions
- ✓ Local Whisper + Piper voice
- ✓ MCP tools on the canvas
- ✓ 80% creator payouts
Explore VibeFuse & harness guides
- Harness Guide
- Free Coding Tools
- AI Coding Agent Desktop
- Free Voice Transcription
- Free Text to Speech
- VibeFuse Product
- Widget Marketplace
- Download Free
- VibeFuse Docs
- Shareable AI Widgets
- Shareable AI Skills
- MCP Tools
- AI Agent Harness
- Harness Engineering
- HyperFrames Video
- Skill Seekers
- Sell AI Skills
- AI Skills Marketplace Compared
- Cursor Alternative
- Local Whisper + Piper
- Custom AI Dashboard
- Community Hub
Claude Code Web UI FAQ
Is there an official Claude Code web UI?
Yes — claude.ai/code. The Remote Control feature (launched February 25, 2026) connects the web interface or the Claude mobile app to a Claude Code session running on your own machine: execution and files stay local, the transcript syncs through Anthropic while connected. It requires a claude.ai login — API keys are not supported.
What are the best self-hosted Claude Code dashboards?
Verified September 2026: Claude Deck (local-only config + sessions + usage dashboard), Clauster (spawns claude remote-control bridges on a homelab or VPS, attach from your phone by QR), claude-pulse (247★, zero-dependency, approve tool calls from your phone), claude-code-dashboard (Windows ConPTY control room with watchdog), claudex (mobile-first, TOTP), and yepanywhere (E2E-encrypted, no accounts).
Can I run Claude Code from my phone?
Three ways: Anthropic's Remote Control (claude.ai/code or the Claude app as a window into a local session — claude.ai login required, one session at a time), a self-hosted dashboard like claude-pulse or Clauster (approve tool calls or reattach from the phone), or the Claude mobile app's cloud sessions, which run on Anthropic infrastructure rather than your machine.
Is a self-hosted Claude Code web UI safe to expose?
Treat any web UI over a shell as a shell on a port. claude-code-web's README warns there is no authentication — anyone reaching the port can control your computer; independent review flags CloudCLI's single-file auth middleware. Keep loopback binds by default, put reverse-proxy auth in front for LAN access, and never expose raw to the internet.
What does Remote Control cost and which plans include it?
It is included with Pro, Max, Team, and Enterprise subscriptions at no extra cost — but only with a claude.ai account login (API keys and setup-tokens cannot establish a session). It is off by default on Team/Enterprise until an owner enables it, and telemetry-disabling flags like DO_NOT_TRACK silently disable it.
How is a widget canvas different from a Claude Code dashboard?
A dashboard is a window or cockpit over terminal sessions — the UI is a chat transcript in a pane, one window per agent. VibeFuse is a desktop canvas where agents live as widgets beside browser, file, video, and notes widgets; you arrange the desk (or ask Vibe Agent + Jarvis to), and marketplace widgets extend it. Windows 10/11, local and offline, free.
Can I make money on widgets the way marketplaces pay for skills?
The self-hosted dashboards and Anthropic's claude.ai/code have no creator payouts at all. VibeFuse runs an open-source marketplace where users make money selling widgets, skills, and styling packs with 80% payouts via Stripe Connect — the monetization layer none of the web-UI projects include.