Tunnelcode — AI skill for Claude Code
Run an AI coding agent on your own machine and drive it from a browser.
How to install Tunnelcode
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open adityadarma/tunnelcode and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Tunnelcode does
Run an AI coding agent on your own machine and drive it from a browser. Pair your phone with a QR code, watch replies stream in, and approve every shell command or file write from wherever you are. Works with OpenCode, Claude Code, Antigravity CLI and Kiro CLI.
Alternatives in AI
- Invisible Dots — Open-source, self-hosted alternative to OpenAI Dots, Meta Muse, Grok Bot, Manus Cue and Claude Cowork 31.8k ★
- BrowserSkill — Let AI agents use your real, logged-in browser without interrupting your work 9.1k ★
- CodeIsland — Real-time AI coding agent status panel in your MacBook notch — live status, approvals & replies for 13 AI tool 2.3k ★
README
TunnelCode
[![gh-workflow-image]][gh-workflow-url] [![npm-image]][npm-url] [![npm-downloads]][npm-downloads] ![][typescript-image] [![license-image]][license-url]
Run an AI coding agent on your own machine and drive it from a browser.
TunnelCode is a bridge between the browser, a server, and a local AI agent. It is not an IDE and not an AI provider. See `PROJECT.md` for the full specification and `DECISIONS.md` for the architecture decisions.
Deploy the server
Docker Compose is how the server is meant to run. You do not need a checkout of this repository: two files and a keypair are the whole of it.
All you need on the host is Docker with the Compose plugin. No Node, no pnpm.
**1. Fetch the two files** into an empty directory:
mkdir tunnelcode && cd tunnelcode
curl -fsSLO https://raw.githubusercontent.com/adityadarma/tunnelcode/master/docker-compose.yml
curl -fsSL -o .env https://raw.githubusercontent.com/adityadarma/tunnelcode/master/.env.example
**2. Create the network** the compose file joins. It is declared external so that a reverse proxy living in its own compose project can reach this container by name. Once per host:
docker network create internet
**3. Generate the push keypair.** The server prints instructions and exits without one, so this is not optional:
docker run --rm --entrypoint node ghcr.io/adityadarma/tunnelcode:latest \
-e "import('./dist/services/web-push.js').then(m => console.log(JSON.stringify(m.generateVapidKeys(), null, 2)))"
**4. Edit `.env`.** Two settings decide whether the container is reachable at all:
- Set
HOST=0.0.0.0. The file ships127.0.0.1, which is right for a checkout and wrong in a container:.envoverrides the image's own value, and loopback inside a container means the published port answers nothing. What controls access here is the port you publish and the firewall in front of it. - Paste both keys from step 3 into
VAPID_PUBLIC_KEYand `VAPID_PRIVATE_
Related Skills
Pocket Homelab AI
Pocket AI — Claude Code and Codex on your server from any browser or phone: resume CLI sessions, watch them wo
Untether
12 Telegram bridge for Claude Code (and 5 other AI agents). Send tasks by voice or text, stream progress, appr
Herdeck
Control panel for AI coding agents running under herdr — on a hardware Stream Deck (Ulanzi D200 or Elgato), a
Cerberus Term
A GUI terminal multiplexer with remote control — native panes that run any CLI (Claude Code, Copilot, Codex, a
Telecode Bridge
Control AI coding agents from Telegram — run opencode, Claude Code & Codex on your own machine: live streaming
Twitch Terminal
Stream your terminal. Watch AI agents work. Like Twitch, for shells.
Related Agents
Community Steward
Triages a public repository's incoming issues and discussions — labels and deduplicates, routes what belongs t
Ade
Agent sessions routinely outlast the machine that started them, and checking a Claude Code or Codex run from a
ML Specialist
Second pair of eyes on machine-learning work — classical ML, deep learning, reinforcement learning, generative