Set-up documentation

Before you add an email routine

An email routine is a job your own AI does in your mailbox on a schedule: a morning brief, replies drafted for the people waiting on you, the emails nobody has answered. You add it from the mailbox's Routines tab, and it becomes one of your AI's own scheduled tasks, run by Claude or ChatGPT under your account.

What each of the 5 does is on the routines page. This one is the screens, in order. You need 3 things first.

• The connection

  The mailbox connected to Claude or ChatGPT

  A routine runs through the same connection you use when you ask your AI about your mail. If that is not set up yet, it takes a few minutes on Claude.ai or from the ChatGPT plugin directory.

• The level

  Read only for most, Draft and file for drafts

  With their starting choices, 4 of the 5 only read. Replies waiting saves drafts, and any routine you set to draft or flag changes your mailbox too, so those need the connection at Draft and file, the level at which the server refuses any send. The Routines tab tells you before you add one the connection cannot run. The 3 levels.

• The plan

  An AI plan with scheduled tasks

  In Claude, scheduled tasks need a paid plan: Pro, Max, Team or Enterprise, per Anthropic's help centre. In ChatGPT the answer is broader, and OpenAI states it precisely.

Scheduled tasks and scheduled task sharing are available to eligible ChatGPT Free, Go, Plus, Pro, Business, Enterprise, and Edu users, and users in eligible ChatGPT for Healthcare workspaces, subject to account and workspace settings.

OpenAI, Scheduled tasks in ChatGPT

So a free ChatGPT account can hold a routine, and that's worth knowing. The same page draws the line that matters for email, though. On Free, a task runs once a day, within a time window such as morning, and an exact time like 07:00 needs a paid plan.

A morning brief that arrives some time in the morning is still a morning brief. A Replies waiting run you wanted done before a 09:00 meeting is a different matter. So pick the plan by the time you need, not by whether the button exists.

Your mailbox has its own allowance as well. A free mailbox has 5 calls a day, which one run of the larger routines can use up, and Pro has 1,000.

Add a routine to Claude or ChatGPT in 6 steps

1. Connect the mailbox to your AI

   If it is not connected yet, do that first, as above. The Routines tab reads which apps the mailbox is connected to and says so in a line above the routines, so you can see before you start whether your AI will be able to reach it.

2. Open the Routines tab

   In the control panel, open the mailbox and select Routines. The 3 steps at the top are the 3 this page walks through: choose, shape, add.

   

3. Choose a routine

   Each card says when it starts, how many calls a run makes and the level it needs. If none of the 4 ready-made ones is the job, choose Your own routine and build it from parts.

4. Shape it, and read what your AI will be told

   Name it, set the days and the time in your own time zone, and make the routine's own choices: who comes first in a morning brief, how far back Replies waiting looks. On the right are the instructions your AI will receive, with the words your choices put there highlighted. Read them once. Nothing is hidden from you, and nothing is sent that is not on that screen.

   

5. Add it to Claude or ChatGPT

   Choose which AI it is written for, then press Add to Claude or Add to ChatGPT. The routine is saved on your mailbox, and your AI opens in a new tab with the whole routine written into its message box. For ChatGPT it starts with $mailbox-mcp, which points ChatGPT at your Mailbox MCP connection rather than at a mail app of its own.

6. Send it, and confirm the task

   Send the message as it stands. Your AI reads it and proposes a scheduled task with the routine's name, days and time. Check them and confirm. This is the step people miss, because it happens in another tab after our part is done, and until it is done nothing is scheduled. The next section says what to look for.

   In ChatGPT

   If ChatGPT asks to connect Gmail, press Not now: your routine reaches your mail through Mailbox MCP instead.

   ChatGPT limits how many scheduled tasks can be on at once. If it says you have reached the limit, pause or delete one of your tasks in ChatGPT, then ask ChatGPT in the same chat to schedule the routine again.

Confirming the scheduled task in your AI

Your AI asks before it schedules anything in your name, which is as it should be, so the last word is always yours. Check that the task it proposes matches the card you just made. The name starts Mailbox MCP: and ends with the mailbox's address in brackets, and the days and the time are the ones you chose.

If the time is wrong, say so in the chat before confirming, rather than fixing it afterwards. A task retimed inside Claude still runs, and we count it by the id it carries wherever it falls in the day. But the card on the Routines tab will describe the schedule you set there.

Every routine begins the same way, and it is worth knowing why. Its first instruction is a call that asks Mailbox MCP which addresses it may use, and in Claude that call carries the routine's own id:

Call list_identities with routine: "3f2a9c1e-0000-4000-8000-000000000001".

That call does 2 jobs. It stops a routine dead if the connection opens a different mailbox from the one it was made for, before anything is read. And it's how we know the run happened: the card can say "Ran today at 07:02" because that call arrived.

A ChatGPT routine is written without the id, because the Mailbox MCP plugin in ChatGPT can't pass one. So a ChatGPT run is matched to its routine by the time it arrives instead.

Check that a routine ran

There are 2 places to look, and they answer different questions. Your AI's own list of scheduled tasks says what it was asked to do and shows what each run wrote. Anthropic puts it in the sidebar:

Review upcoming and past runs by clicking "Scheduled" in the left sidebar on any surface.

Anthropic, Schedule recurring tasks in Claude Cowork

So the list is wherever you use Claude, which is also where you read the brief itself. ChatGPT has the equivalent, a Scheduled page listing every task and its results.

Cowork, in the article's title, is the part of Claude that carries out tasks for you rather than chatting. Anthropic is folding it into Claude itself, so you may not see the name.

Neither app can tell you about a run that never started, because a run that doesn't start writes nothing to look at. That's the gap the Routines tab covers. Each saved routine's card says whether it ran, when, and how many calls it made, in the routine's own time zone.

We've run a scheduled morning brief in Claude and in ChatGPT side by side on our own demo mailbox since 2 October 2026, and read what each run said in its own app's list.

Change, pause or remove a routine

• Change

  Edit, then Update in Claude or ChatGPT

  On the routine's card, press Edit, change what you want, and press Update in Claude or Update in ChatGPT. The change is saved on the card, and your AI is asked to update the task it already has rather than make a second, so its run history stays. Duplicate starts a new routine from a copy.

• Pause

  In your AI's Scheduled list

  Pausing is your AI's, on the same Scheduled list. If you pause a routine there, its card here will say "Did not run" after its next due time, which is true. Remove it here as well if you do not mean to start it again.

• Remove

  Remove, then ask your AI to stop its task

  Press Remove on the card and confirm. The routine and its card leave the Routines tab, but your AI keeps running its scheduled task until you stop it there. So the next screen has a button that asks it to.

  Claude deletes the task. ChatGPT can't delete one from a chat, so the button asks it to switch the task off, which pauses it. To remove it altogether, delete it on ChatGPT's Scheduled page, where it shows as Paused. The message names the task and tells your AI to change nothing if no such task is found.

When a scheduled routine does not run

If a run doesn't come, the card says so in words, with when it was due, when it last ran and what to check. It waits 2 hours after the due time in Claude, and 6 in ChatGPT, before it calls the run missed, because both apps can start a run a little late. When the routine runs again, the card says that too.

Whether it ran is on the card. What each run found is in your AI's own Scheduled list, as it is for any task it runs.

The causes are nearly always one of 4, and the first 2 are the common ones:

1. The task was never confirmed. Step 6 above. Once its first due time has passed with no run, the card says "Not started yet".
2. The task was paused or deleted in your AI. OpenAI notes that inactive tasks may pause automatically, so a routine nobody has looked at for a while can stop on its own. If it is not in the Scheduled list any more, press Add to Claude again or Add to ChatGPT again on the card.
3. The AI lost its connection to the mailbox. Reconnect it the way you first connected it; the routine needs nothing else.
4. The plan changed. A Claude plan that is no longer paid has no scheduled tasks, and on ChatGPT Free a task runs once a day, within a time window, rather than at the time you set.

Where to go next

What each routine does, with the exact instructions it sends, is on its own page: Morning brief, Replies waiting, Waiting on, Bills and invoices and Your own routine. Why a routine is built the way it is, and what it may do in your mailbox, is on the routines page.

Read 2 October 2026 The Claude and ChatGPT steps were read off Anthropic's and OpenAI's help centres, each linked where it is quoted, and the Routines tab's screens off our own control panel at version 2.0.0, on that day. The screen names in Claude and ChatGPT are theirs and may move: scheduled tasks are changing quickly in both. If it has changed since, tell us and we will read it again.