🇺🇸 USA · Claude Remote Dispatch

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Fire-and-forget remote Claude — trigger an agent task that runs in a cloud sandbox without keeping your local Claude Code session attached. The “I’m leaving for lunch, come back to results” mode.


Front-matter facts

FieldValue
VendorAnthropic (San Francisco, USA)
Country / origin🇺🇸 USA
Recommended for Australian users?✅ Yes
Privacy summarySame Anthropic data-handling; ephemeral sandbox state
Free tierLimited via Claude Pro / Max
Paid tiersBundled with Claude Pro / Max / Team / Enterprise
First released2025-26 — rolled out alongside Claude Remote Desktop
Last reviewed2026-06-26
Official sitehttps://docs.claude.com/en/docs/claude-code

What it is

Claude Remote Dispatch is Claude Code’s fire-and-forget remote agent trigger. Unlike Claude Remote Desktop (which gives you an interactive remote Claude session), Remote Dispatch lets you submit a task and disconnect — Claude runs the task in a cloud sandbox without your local session attached, and reports back when complete via notification, PR, or email.

Mental model:

  • Remote Desktop = “I want to drive a Claude in the cloud interactively” (like SSH-ing into a remote dev machine)
  • Remote Dispatch = “I want to hand Claude a task and walk away” (like submitting a job to a queue and getting a result later)

Practical use:

  • Submit a long refactor at end of workday — wake up to a PR
  • Dispatch parallel tasks across many repositories — get a summary later
  • Schedule recurring tasks (e.g., “every Monday, summarise last week’s PRs”)

What you’d use it for

  • Background tasks that you don’t want occupying your local Claude session
  • End-of-day refactor submissions — Claude works overnight (in AUS terms, that’s overnight in the US too — but the sandbox runs whenever)
  • Multi-repo operations — dispatch one task per repo
  • Recurring scheduled tasks — combined with cron-like triggers
  • CI-style automation — dispatch on git events
  • “Set it and forget it” research — long-running exploration tasks

How to use it

  1. From Claude Code (or via the Anthropic API directly):
    • Submit a dispatch: /dispatch <task-description> --repo <path> --notify-when-done
    • Configure how you want to be notified (email, GitHub PR, Slack message, etc.)
  2. Claude provisions a cloud sandbox; runs the task; reports back
  3. Your local Claude Code session is free to do other things in parallel

For developers, the Claude Agent SDK exposes Remote Dispatch programmatically — you can trigger from your own code, CI pipelines, scheduled cron, etc.


How it compares to alternatives

CapabilityClaude Remote DispatchClaude Remote DesktopOpenAI Codex (cloud)GitHub Actions + Claude
Fire-and-forgetYes (designed for it)Possible but interactiveYesYes
Interactive supervisionNo (by design)YesLimitedNo
Scheduled / triggeredYesNoManualVia cron / events
Bundled with Claude subscriptionYesYesChatGPT subscriptionPay-per-API + Actions minutes
Best forBackground async workInteractive cloud sessionsCloud agent tasksCustom CI workflows

For background async work, Remote Dispatch is the most natural fit within the Anthropic ecosystem.


Privacy / data handling

  • Same Anthropic posture; ephemeral sandboxes destroyed after task completion
  • Notification methods (email / Slack / GitHub) involve standard third-party services — they handle the notification payload
  • For sensitive code, use Team / Enterprise tier where no-training is contractual

Recent changes

  • 2026: Remote Dispatch and Remote Desktop matured together; coordination UX improved
  • Late 2025: Initial rollout

Gotchas

  • Don’t dispatch tasks that need your judgment mid-flow — Remote Dispatch is for tasks where Claude can complete autonomously. If you need supervision, use Remote Desktop or local Claude Code instead.
  • Notification reliability matters — verify the notification path before relying on it for important results
  • Credential management — sandboxes need git / npm / cloud credentials configured; use scoped tokens
  • Cost scales with run-time — long-running dispatches consume more API tokens
  • Failed dispatches need investigation — when a dispatched task fails, you only see the final report; harder to debug than interactive sessions

See also


Sources