Remote Control: Drive Claude From Anywhere

INPUT LAYER
  Slack DM  →  bot webhook  ┐
  Telegram  →  bot webhook  ├──→  Local relay server  →  Claude Code CLI
  SMS       →  Twilio hook  ┤                               ↓
  Voice     →  iOS Shortcut ┘                          Result handler

OUTPUT LAYER
  Result handler  →  reply via original channel (Slack / Telegram / SMS)
               or  →  file saved to disk  (for large outputs)
               or  →  silent  (fire-and-forget tasks)

// relay-server.mjs
import express from "express"
import { execSync } from "child_process"

const app = express()
app.use(express.json())

app.post("/trigger", (req, res) => {
  const { text, secret } = req.body
  if (secret !== process.env.WEBHOOK_SECRET) return res.sendStatus(401)

  const result = execSync(
    `claude --message ${JSON.stringify(text)} --output-format text`
  )
  res.json({ result: result.toString() })
})

app.listen(3333)

On the train or metro

You had a session going, left your laptop at the office, and realise on the train that you forgot to push a critical fix. Scan the QR code you generated before leaving, continue from where you stopped, push the commit, done.

# Before leaving desk
/rc
# Scan QR on phone → continue session in mobile browser

1. 1

   One-time URL

   Each /rc generates a unique, short-lived URL. It expires after a set time or when the session ends. No permanent open port left behind.

2. 2

   Local tunnel only

   The connection runs through a secure local tunnel. Your code never leaves your machine through an unsecured channel.

3. 3

   Hook-level guardrails

   Your existing bash-safety.sh and file-protection.sh hooks still run. Remote triggers go through the exact same permission layer as desktop commands.

4. 4

   Avoid public WiFi for sensitive sessions

   If you are working with credentials, production systems, or confidential code, use a mobile hotspot instead of public WiFi when connecting remotely.