Claude Dispatch Not Working: What to Check First

Claude Dispatch may not work because your account lacks access, the desktop is unavailable, or the task is waiting for an answer or permission. Start by checking eligibility: Anthropic says Dispatch is unavailable to new users in its Dispatch help page↗ (opens in a new tab), so a missing feature is not necessarily a connection failure.

fenbs can keep the original task and last confirmed progress available while you diagnose the interruption. That record helps you restart from evidence instead of asking another session to repeat everything.

What should you check before restarting anything?

1. Identify the symptom: Dispatch is missing, messages do not arrive, or a particular task stopped progressing. These need different checks.
2. Confirm the signed-in account on both devices. Check that the task you are inspecting belongs to that account and the intended desktop.
3. For an existing desktop Dispatch task, keep the computer awake, the desktop app open, and both devices online.
4. Read the active session for a question, permission request, error, or completed output before resending instructions.
5. Write down the last action you can verify, then retry only the unfinished part after addressing the cause.

Avoid changing several settings at once. If you update an app, reconnect a device, and rewrite the task together, you will not know which change helped. Keep a short troubleshooting record: symptom, check, observed result, and next check. You can stop once the original task is working and its output is verified.

Is Dispatch missing or is an existing connection failing?

A missing feature and a broken run are different problems. Check the vendor instructions for account availability before reinstalling applications. If your account never had Dispatch, repeated sign-ins or permission changes are unlikely to make an unavailable feature appear. Use the workflow Anthropic currently offers to your account.

If Dispatch worked before, compare the account and device with your last successful use. A work account on the phone and a personal account on the desktop are easy to confuse. Confirm the account identity in the app rather than relying on a familiar display name or the appearance of the home screen.

For the normal setup flow, see how to use Claude Dispatch. Keep troubleshooting focused on the first check that fails; completing setup again does not explain a task that has already produced half its output.

Can the task reach the computer and files it needs?

Dispatch desktop work needs a reachable computer, while newer Cowork cloud sessions have a different execution model. Anthropic explains which work depends on your computer in its web, desktop, and mobile guide↗ (opens in a new tab). Identify the actual task type before treating a closed laptop as the explanation for every interruption.

Test the smallest relevant dependency. If the assistant cannot read a local document, confirm that the document is still present and accessible from the desktop. If it can read one folder but not another, examine that folder’s access instead of assuming the entire connection is broken.

A moved or renamed file can make an old instruction stale. Give the assistant the current location and ask it to confirm that it found the intended input before doing more work. Do not share passwords or access tokens in a troubleshooting message; describe the affected connection without copying its credentials.

Save the corrected source location with the task. The next session can then use the fenbs handoff record instead of inheriting the same broken instruction.

Is the assistant waiting for you rather than stuck?

Open the existing task and read its latest request. It may need a choice between two files, permission to access an app, or clarification of the intended output. A missing phone notification does not establish that no question was asked. Inspect the task itself before changing the connection.

Permissions should match the action you want. Anthropic describes per-app access in its computer-use documentation↗ (opens in a new tab). Granting broader access to make a warning disappear can change the task’s scope without fixing the underlying problem.

If the request is understandable and necessary, respond and watch the next observable step. If it is unrelated, restate the boundary and ask for a narrower route. “Finish the draft without sending it” gives a clearer recovery instruction than “continue however you need to.”

How do you retry without duplicating changes?

Check the destination before repeating an action that changes something. Look for a saved file, updated spreadsheet, sent message, or created record. If you cannot verify whether an action happened, say so explicitly and ask the assistant to inspect that state before attempting it again.

Do not restart the full task.
First inspect the existing output and report what is already complete.
Last confirmed result: the draft agenda exists in the review folder.
Unverified: whether the action-owner table was added.
Next: check that table and complete only the missing rows.
Do not send messages or replace the source notes.

A clean recovery can leave an unfinished task on purpose. If a source is inaccessible, preserve the checked draft and name the missing source rather than asking the assistant to fill gaps from guesses. Our guides to Cowork failures and Claude Code failures cover problems inside those working environments.

How to do this with fenbs

1. Create a free account and add the project whose handoff was interrupted.
2. Add or update its task with the intended result, current source locations, and the error you observed. Put recovery steps in the plan.
3. Record what you verified and what remains uncertain. Save Where we left off with the first check the next session should perform.
4. Connect a supported assistant and ask it to read the project context before continuing. Update the task and save progress after the recovery check.

fenbs does not repair Dispatch connections or display a live mirror of its sessions. For a one-time connection issue, the vendor help instructions may be sufficient. It is useful when the interruption would otherwise lose the work’s current state across assistants.

Use fenbs AI context to preserve a checked continuation, and explore the AI handoff library for other transitions. The next assistant should be able to name both the unfinished action and the evidence it must inspect before acting.