Claude Dispatch

Claude Dispatch is a feature within Anthropic's Claude platform that lets users remotely control a Claude Cowork session running on their desktop computer from a mobile device, assigning tasks from a phone and returning later to finished work.^[1][15] Launched on March 17, 2026, as a research preview, Dispatch creates a single persistent conversation thread that spans the Claude mobile app (iOS or Android) and the Claude Desktop app (macOS or Windows), so a user can start a task on the phone and pick up the completed output on the computer.^[15][16] All processing runs locally on the user's desktop, meaning files and data never leave the machine during task execution.^[1][2]

Anthropic engineer Felix Rieseberg introduced the feature on X, describing it as "one persistent conversation with Claude that runs on your computer. Message it from your phone. Come back to finished work."^[15] 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.^[16][2]

Background and Context

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.^[3] 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.^[3]

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.^[1][16]

The feature was announced on March 17, 2026, by Anthropic engineer Felix Rieseberg, who posted on X that Dispatch is "one persistent conversation with Claude that runs on your computer. Message it from your phone. Come back to finished work."^[15] On March 23, 2026, Anthropic published a broader blog post pairing Dispatch with the launch of computer use capabilities as a research preview, describing the workflow as one where a user can "assign Claude a task on your phone, turn your attention to something else, then open up the finished work on your computer."^[2]

Cowork itself continued to expand after Dispatch launched. On July 7, 2026, Anthropic brought Cowork to the web and mobile for Max subscribers, adding cloud-backed sessions that can keep working even when the user's computer is closed.^[18] These cloud sessions are distinct from Dispatch, which remains a remote control for a session executing on the user's own desktop rather than in the cloud.^[18][2]

How does Claude Dispatch work?

Architecture

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.^[1][5][6]

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.^[1][2]

Persistent Conversation Thread

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.^[1][14][16]

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.

Setup Process

Setting up Dispatch takes approximately two minutes and requires no API keys or complex configuration:^[1][11][17]

Scroll sideways for more →

Once paired, tapping the Dispatch entry in the mobile app's sidebar instantly connects to the desktop Cowork session. If a user is already signed in to the same Claude account on both devices, the conversation syncs automatically after setup.^[10][15][17]

System Requirements

Scroll sideways for more →

What can Claude Dispatch do?

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.^[1]

File and Document Operations

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.^[1]

Communication and Research

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.^[1]

Presentation and Report Generation

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.^[1]

Computer Use Integration

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. In Anthropic's words, Claude can "point, click, and navigate what's on your screen to perform the task itself."^[2] 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.^[2]

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.^[2][9]

Recurring and Scheduled Tasks

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.^[1][2]

Is Claude Dispatch secure?

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.^[9]

Local Execution and Data Privacy

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.^[1][2]

Folder-Level Permissions

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.^[9][12]

Scroll sideways for more →

Human-in-the-Loop Controls

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.^[9][6]

Cowork also offers an "Ask before acting" mode, which Anthropic recommends for tasks that touch sensitive files, accounts, or sites, and for situations where a mistake would be hard to undo.^[9] 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 "complex tasks sometimes need a second try, and working through your screen is slower than using a direct integration," recommending that users start with trusted applications and avoid exposing sensitive data during the preview period.^[2][9]

Push Notifications

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.^[1][4]

Kill Switch

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.^[9]

Known Security Considerations

Anthropic has identified several areas of caution for Dispatch users:

• Prompt injection risk: If Claude encounters manipulated content during task execution (for example, malicious instructions embedded in a file it reads), those instructions could potentially influence its behavior. Anthropic runs content classifiers that "scan all untrusted content entering Claude's context and flag potential injections," and advises granting access only to files and connectors the user is comfortable with Claude acting on autonomously.^[9]
• Computer use risk: When Claude uses computer use mode to interact with on-screen applications, it operates outside the normal sandbox. As Anthropic puts it, "computer use has no sandbox between Claude and what's on your screen," so browser interactions could expose Claude to external content that might contain adversarial prompts.^[9]
• No undo mechanism: Actions performed by Dispatch (file modifications, deletions, emails sent) cannot be automatically reversed, though Cowork requires explicit permission before permanently deleting a file. Users should verify the scope of permissions before granting broad access.^[9]
• Audit and compliance gap: Cowork activity, including Dispatch actions, is not captured in Anthropic's Compliance API at this time. Team and Enterprise administrators can, however, stream Cowork events to their own SIEM and observability tools through OpenTelemetry. Anthropic advises against scheduling tasks that access sensitive files or take actions that are difficult to undo.^[9]

What are Claude Dispatch's limitations?

As a research preview, Dispatch has several known limitations:

Scroll sideways for more →

Anthropic itself frames the reliability caveat plainly, noting that "complex tasks sometimes need a second try, and working through your screen is slower than using a direct integration."^[2] The single-thread and desktop-awake constraints are what distinguish Dispatch from the cloud-backed Cowork sessions Anthropic later shipped on web and mobile, which continue running even when the user's computer is closed.^[18][4]

How does Claude Dispatch differ from OpenClaw?

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 late January 2026 and then to OpenClaw days later. By March 2026, OpenClaw had become one of the fastest-growing repositories in GitHub history by star count, surpassing 250,000 stars within roughly its first three months and continuing to climb past 340,000 stars by mid-2026. Steinberger announced in February 2026 that he would join OpenAI and transfer the project to an open-source foundation.^[13][8]

OpenClaw and Claude Dispatch address overlapping use cases (remote AI task execution) but take fundamentally different approaches.^[8]

Scroll sideways for more →

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.^[7][8]

As noted by The New Stack, "Dispatch isn't replacing OpenClaw because they serve different users."^[8]

What is Claude Dispatch used for?

Dispatch is best suited for asynchronous workflows where the user does not need to supervise Claude's work in real time.^[1]

Morning Briefings

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.

Report Generation

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.

File Organization

Dispatch can sort and organize large directories of files, rename documents according to conventions, and move items into appropriate folders based on content analysis.

Development Workflows

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.^[2]

Meeting Preparation

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.

How much does Claude Dispatch cost?

Claude Dispatch is included at no additional cost for subscribers on the following plans:^[16][2]

Scroll sideways for more →

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.^[16] Team and Enterprise plan access has not been announced as of March 2026, though Cowork itself already supports those tiers.^[3]

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.^[1]

Relationship to Claude Code Remote Control

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.^[1]

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.^[2]

Reception

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.^[4]

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.^[4]

Multiple publications, including Forbes 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.^[16][8]

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 Compliance API coverage and the risks associated with computer use mode were flagged as areas requiring attention before enterprise adoption.^[9][6]

See Also

• Anthropic
• Claude
• Claude Code
• AI Agents
• OpenClaw
• Prompt Engineering

References

1. "Assign tasks to Claude from anywhere in Cowork." Claude Help Center, Anthropic. https://support.claude.com/en/articles/13947068-assign-tasks-to-claude-from-anywhere-in-cowork ↩
2. "Put Claude to work on your computer." Claude Blog, Anthropic, March 23, 2026. https://claude.com/blog/dispatch-and-computer-use ↩
3. "Introducing Cowork." Claude Blog, Anthropic. https://claude.com/blog/cowork-research-preview ↩
4. "Hands-On with Claude Dispatch for Cowork." MacStories. https://www.macstories.net/stories/hands-on-with-claude-dispatch-for-cowork/ ↩
5. "Claude Dispatch Explained: What It Is and How It Works." LowCode Agency. https://www.lowcode.agency/blog/claude-dispatch-explained ↩
6. "Claude Dispatch: The Constraints, the Security Model, and What Comes Next." DEV Community. https://dev.to/ji_ai/claude-dispatch-the-constraints-the-security-model-and-what-comes-next-hji ↩
7. "Claude Dispatch vs OpenClaw: Which AI Agent Tool Should You Use?" Techloy. https://www.techloy.com/claude-dispatch-vs-openclaw-which-ai-agent-tool-should-you-use/ ↩
8. "Anthropic's response to the AI tool that caused lines around the block in Shenzhen." The New Stack. https://thenewstack.io/claude-dispatch-versus-openclaw/ ↩
9. "Use Cowork safely." Claude Help Center, Anthropic. https://support.claude.com/en/articles/13364135-use-cowork-safely ↩
10. "Claude Dispatch: Control Cowork From Your Phone (Setup + First Look)." FindSkill.ai. https://findskill.ai/blog/claude-dispatch-remote-cowork/ ↩
11. "Anthropic Adds Claude Cowork Dispatch for Remote Desktop Control From Your Phone." Geeky Gadgets. https://www.geeky-gadgets.com/claude-cowork-dispatch-guide/ ↩
12. "Claude Dispatch Setup and Permissions in Claude Apps." Geeky Gadgets. https://www.geeky-gadgets.com/claude-dispatch-security-permissions/ ↩
13. "OpenClaw." Wikipedia. https://en.wikipedia.org/wiki/OpenClaw ↩
14. "Claude Cowork Dispatch Launch: Persistent Cross-Device AI Workflow Preview for 2026 Productivity." Blockchain News. https://blockchain.news/ainews/claude-cowork-dispatch-launch-persistent-cross-device-ai-workflow-preview-for-2026-productivity ↩
15. Felix Rieseberg. Post announcing Claude Dispatch. X (formerly Twitter), March 17, 2026. https://x.com/felixrieseberg/status/2034005731457044577 ↩
16. Schmelzer, Ron. "Anthropic Update Lets You Control Claude Cowork With Your Phone." Forbes, March 20, 2026. https://www.forbes.com/sites/ronschmelzer/2026/03/20/claude-dispatch-lets-you-control-claude-cowork-with-your-phone/ ↩
17. "Claude Cowork Dispatch 101: Remote Control for Desktop AI." DataCamp. https://www.datacamp.com/tutorial/claude-cowork-dispatch ↩
18. "The coding agent wars are spilling into the rest of the office with Claude Cowork." TechCrunch, July 7, 2026. https://techcrunch.com/2026/07/07/the-coding-agent-wars-are-spilling-into-the-rest-of-the-office-claude-cowork/ ↩