Claude Dispatch is a feature within Anthropic's Claude platform that enables users to remotely control a Claude Cowork session on their desktop computer from a mobile device. Launched on March 17, 2026, as a research preview, Dispatch creates a single persistent conversation thread between the Claude mobile app (iOS or Android) and the Claude Desktop app (macOS or Windows), allowing users to assign tasks from their phone and return to completed work on their computer. All processing runs locally on the user's desktop, meaning files and data never leave the machine during task execution.
Dispatch was initially available to Claude Max subscribers, with access extended to Claude Pro subscribers within days of launch. The feature is included at no additional cost for both tiers.
Claude Cowork, Anthropic's desktop agent for knowledge work, launched as a research preview on January 12, 2026, for Claude Max subscribers on macOS. It expanded to Pro subscribers on January 16, to Team and Enterprise plans on January 23, and to Windows on February 10. Cowork allows Claude to read, edit, and create files within user-specified folders, plan multi-step workflows, and execute tasks autonomously on the local machine.
Before Dispatch, interacting with Cowork required sitting at the computer where the Claude Desktop app was running. Users who wanted to assign a task and walk away had no built-in mechanism to monitor progress or provide approvals from another device. Dispatch was designed to solve this limitation by turning the Claude mobile app into a remote control interface for an active Cowork session.
The feature was announced on March 17, 2026, by Anthropic engineers Felix Rieseberg and Boris Cherny. In the announcement, they described Dispatch as enabling "one persistent Claude conversation that runs locally on your computer and can be messaged from your phone, so you can return to finished work." On March 23, 2026, Anthropic published a broader blog post pairing Dispatch with the launch of computer use capabilities as a research preview, further extending what Dispatch can control remotely.
Dispatch operates on a local execution model. When a user sends an instruction from the Claude mobile app, that instruction is routed through Anthropic's servers to the Claude Desktop app running on the user's computer. The desktop app then executes the task using local files, installed plugins, configured connectors, and any other resources available to the Cowork session. Results are synced back to the mobile app for the user to review.
The key architectural principle is that all computation happens on the user's desktop. Files are never uploaded to Anthropic's cloud infrastructure during task execution. Anthropic's servers serve as a relay layer for instructions and state synchronization, but the actual work (file reading, editing, code execution, document generation) is performed locally.
Unlike standard Claude chat sessions, which are separate and stateless between devices, Dispatch maintains a single continuous conversation thread that is accessible from both the phone and the desktop. Messages sent from either device appear in the same thread, and Claude retains context across the entire conversation history. This eliminates the need to re-explain ongoing projects or re-establish context when switching between devices.
The persistent thread also means that Claude's memory feature carries over. If Claude learned about a user's preferences, file organization, or project structure in a previous Cowork session, that context remains available in the Dispatch thread.
Setting up Dispatch takes approximately two minutes and requires no API keys or complex configuration:
| Step | Action |
|---|---|
| 1 | Download or update the Claude Desktop app to the latest version (macOS or Windows x64) |
| 2 | Download or update the Claude mobile app to the latest version (iOS or Android) |
| 3 | Open the Cowork tab in the Claude Desktop app |
| 4 | Click "Dispatch" in the left panel |
| 5 | Select "Get started" and configure file access and keep-awake settings |
| 6 | A QR code appears on the desktop screen |
| 7 | Open the Claude mobile app, tap "Pair with your desktop," and scan the QR code |
| 8 | The Dispatch conversation thread appears on both devices |
Once paired, tapping the Dispatch entry in the mobile app's sidebar instantly connects to the desktop Cowork session.
| Requirement | Details |
|---|---|
| Desktop OS | macOS (Apple Silicon M1 or later for Cowork tab) or Windows x64 |
| Desktop App | Latest version of Claude Desktop, must remain open and running |
| Mobile App | Latest version of Claude mobile app (iOS or Android) |
| Subscription | Claude Pro ($20/month) or Max ($100-$200/month) plan |
| Connectivity | Active internet connection on both devices |
| Desktop State | Computer must remain awake; sleeping or closing the lid halts all task execution |
Dispatch can execute any task that a Cowork session has permission to perform. Because it inherits the full capability set of the desktop Cowork environment, the range of possible tasks is broad.
Users can instruct Claude to search for specific files, read and summarize documents, create new files, edit existing content, organize folders, and compile reports from local data. For example, a user might send a message from their phone asking Claude to "pull together a summary of this week's sales data from the spreadsheets in my Reports folder," and Claude will locate the relevant files, process them, and produce a formatted document ready for review.
With the appropriate connectors configured, Dispatch can search through email inboxes and Slack channels, draft briefing documents, summarize meeting notes, and pull information from connected services like Notion, Google Drive, and other integrated platforms.
Dispatch can generate structured outputs such as slide decks (.pptx files), formatted reports, and data summaries. Claude works through these tasks step by step within the Cowork session, producing downloadable files that the user can retrieve from their phone.
As of March 23, 2026, Dispatch also works with Claude's computer use capabilities. When Claude does not have a direct tool integration for a particular application, it can interact with the desktop by pointing, clicking, scrolling, and navigating on-screen elements. This means that from a phone, a user can instruct Claude to perform actions in applications that do not have dedicated plugins, such as navigating a web browser, using desktop software, or interacting with development tools.
Claude prioritizes direct tool integrations first and falls back to computer use (screen interaction) only when needed. The computer use feature operates outside the normal sandbox, which introduces additional risk that users should be aware of.
Dispatch supports scheduling recurring tasks, such as reviewing open pull requests every morning, auditing dependencies on a weekly basis, or checking for CI failures overnight. These automated tasks run while the computer is awake and the Claude Desktop app is open.
Anthropic built Dispatch with several safety mechanisms, consistent with the company's emphasis on AI safety. Because Dispatch enables remote commands that trigger real actions on a local computer, the security model is a central design consideration.
All task execution occurs on the user's machine. Files are processed locally and are not transmitted to Anthropic's servers during task execution. This means sensitive documents (legal files, financial records, medical data, credentials) stay on the user's hardware throughout the workflow. Anthropic's infrastructure relays instructions and synchronizes conversation state but does not access or store the content of local files.
Cowork (and by extension Dispatch) uses an explicit, folder-scoped access model. During setup, users select specific folders that Claude is permitted to read, write, and modify. Claude cannot access files outside these designated directories. Anthropic recommends creating a dedicated working folder rather than granting access to broad parent directories that may contain unrelated sensitive material.
| Permission Type | Description |
|---|---|
| Folder access | Claude can only interact with files inside user-selected folders |
| Application access | Claude asks for permission before accessing each new application (in computer use mode) |
| Plugin access | Only plugins and connectors the user has explicitly installed and configured are available |
| Blocked applications | Some applications are blocked by default to protect sensitive data |
Dispatch incorporates approval prompts that pause the agent before performing potentially destructive actions. When Claude is about to execute an operation such as deleting files, moving large directories, or performing other irreversible changes, it pauses and sends a push notification to the user's phone requesting approval. The user must explicitly allow the action before Claude proceeds.
This human-in-the-loop design is intended to prevent unintended consequences from remote instructions. However, as with any research preview, the system is not without limitations. Anthropic acknowledges that "computer use is still early" and that "Claude can make mistakes," recommending that users start with trusted applications and avoid exposing sensitive data during the preview period.
Dispatch sends push notifications to the user's mobile device in two scenarios: when a task is completed and results are ready for review, and when Claude requires approval to proceed with a sensitive action. Notification reliability has been noted as an area that still needs improvement during the research preview phase.
Users can immediately halt all Dispatch activity by closing the Claude Desktop app, putting the computer to sleep, or powering it off. This provides a straightforward emergency stop mechanism.
Anthropic has identified several areas of caution for Dispatch users:
As a research preview, Dispatch has several known limitations:
| Limitation | Details |
|---|---|
| Desktop must stay awake | If the computer sleeps or the Claude Desktop app is closed, all task execution stops. Dispatch is a remote control, not cloud computing. |
| Single conversation thread | Only one persistent Dispatch thread is available at a time. Users cannot run multiple parallel Dispatch conversations. |
| One task at a time | Dispatch processes tasks sequentially. It cannot handle multiple concurrent requests. |
| Reliability | Early reviewers report roughly 50% success rates on complex, multi-step tasks. Information retrieval tasks succeed more consistently than system-level operations. |
| Cross-app limitations | Actions that span multiple applications (such as composing an iMessage, running AppleScript commands, or managing Safari tabs) frequently fail in the current preview. |
| No offline capability | Both devices require active internet connections. Dispatch does not function on local networks alone. |
| macOS Cowork restriction | The Cowork tab on macOS requires Apple Silicon (M1 or later). Intel Macs can use Chat and Code tabs but not Cowork. |
OpenClaw is an open-source AI agent platform that was originally developed by Austrian developer Peter Steinberger under the name Clawdbot in November 2025. After trademark complaints from Anthropic, the project was renamed to Moltbot in January 2026 and then to OpenClaw days later. By March 2026, OpenClaw had become the fastest-growing repository in GitHub history by star count, accumulating over 247,000 stars. Steinberger announced in February 2026 that he would join OpenAI and transfer the project to an open-source foundation.
OpenClaw and Claude Dispatch address overlapping use cases (remote AI task execution) but take fundamentally different approaches.
| Feature | Claude Dispatch | OpenClaw |
|---|---|---|
| Architecture | Instructions routed through Anthropic servers, executed locally on desktop | Runs entirely locally with optional cloud API connections; can operate fully offline with local models |
| Supported platforms | Claude mobile app only (iOS/Android to macOS/Windows) | WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Microsoft Teams, and more |
| AI model | Claude only | Supports Claude, GPT-4, Gemini, DeepSeek, and local models |
| Pricing | Included with Claude Pro ($20/month) or Max ($100-$200/month) | Free and open-source (MIT license); API costs vary by usage ($5-$20/month typical) |
| Setup complexity | Two-minute QR code pairing; no technical configuration required | Requires technical setup; command-line installation and configuration of API keys and messaging integrations |
| Security defaults | Built-in sandboxing, folder permissions, and approval prompts | No strong default safeguards; security depends on user configuration; researchers have flagged vulnerabilities including zero-click exploits |
| Open source | Proprietary, closed system | Fully open-source under MIT license |
| Scheduling | Supports recurring tasks | Supports scheduled tasks and webhooks |
| Multi-model support | No | Yes |
| Desktop OS support | macOS, Windows x64 | macOS, Windows, Linux |
The two tools serve different user profiles. Dispatch is designed for users already within the Claude ecosystem who want a frictionless, secure mobile-to-desktop workflow without technical setup. OpenClaw appeals to technically proficient users who want flexibility in choosing AI models, messaging platforms, and custom automation configurations.
As noted by The New Stack, "Dispatch isn't replacing OpenClaw because they serve different users."
Dispatch is best suited for asynchronous workflows where the user does not need to supervise Claude's work in real time.
A user can send a Dispatch message from their phone while commuting, asking Claude to compile a briefing from overnight emails, Slack messages, and calendar updates. By the time the user arrives at the office, a summary document is ready on the desktop.
Knowledge workers can point Dispatch at a folder of spreadsheets or data files and request a formatted weekly report. Claude processes the files locally and produces a downloadable document.
Dispatch can sort and organize large directories of files, rename documents according to conventions, and move items into appropriate folders based on content analysis.
With computer use enabled, developers can use Dispatch to trigger code reviews, run test suites, check CI/CD pipelines, and submit pull requests from their phone. This is particularly useful for monitoring long-running processes while away from the desk.
Before a meeting, a user can dispatch Claude to pull together relevant documents, summarize recent project updates, and prepare talking points, all from a quick message sent during a walk between conference rooms.
Claude Dispatch is included at no additional cost for subscribers on the following plans:
| Plan | Monthly Price | Dispatch Access |
|---|---|---|
| Claude Pro | $20/month | Available (added within days of March 17, 2026 launch) |
| Claude Max 5x | $100/month | Available from March 17, 2026 launch |
| Claude Max 20x | $200/month | Available from March 17, 2026 launch |
| Claude Team | Varies | Not yet announced |
| Claude Enterprise | Varies | Not yet announced |
| Free tier | Free | Planned for later in 2026 |
Max subscribers received access first on the March 17 launch date. Pro subscribers began receiving access within days, with widespread availability confirmed by approximately March 22, 2026. Team and Enterprise plan access has not been announced as of March 2026, though Cowork itself already supports those tiers.
Dispatch requires no separate purchase or add-on. It is a feature within the existing Claude Desktop and mobile apps, accessed through the Cowork interface.
Claude Dispatch for Cowork is conceptually similar to the Remote Control feature in Claude Code, Anthropic's command-line coding agent. Claude Code's Remote Control, which preceded Dispatch, allows developers to manage terminal-based Claude Code sessions from a mobile device. Dispatch extends this same phone-to-desktop paradigm to Cowork's broader knowledge-work capabilities.
As of March 23, 2026, Dispatch is available in both Cowork and Claude Code, meaning the same persistent conversation can leverage both knowledge-work tools and development tools depending on the task at hand.
Early reception of Claude Dispatch has been mixed. Reviewers have praised the concept and the simplicity of setup, while noting that the research preview's reliability leaves room for improvement.
MacStories described Dispatch as having "about a 50/50 shot whether what you try will work," noting that information retrieval tasks succeeded consistently but cross-application actions frequently failed. The review highlighted that Dispatch is "slow" compared to direct interaction with Cowork on the desktop.
Multiple publications, including Latent Space and The New Stack, framed Dispatch as Anthropic's response to the viral growth of OpenClaw, which had demonstrated strong demand for mobile-controlled AI agents.
The security model received generally positive commentary, with reviewers noting that local execution and folder-scoped permissions provide stronger default protections than many competing approaches. However, the lack of audit logging and the risks associated with computer use mode were flagged as areas requiring attention before enterprise adoption.