Connecting Claude to your LifeOps

What it isA connected Claude, and the two Claudes it is not

Three different things in this system are called Claude, and they reach different data.

• A connected Claude is any Claude conversation you have joined to your instance. This page is about that one. It reads and writes your real data through seventeen named tools and nothing else.
• Chat is the speech bubble inside the app. It answers about your data too, and the user's guide covers it. It runs on the Anthropic key stored in Settings and needs no connecting.
• Your Claude is the Claude Code session that builds the app. It changes code, ships releases and runs deploys. Building with Claude Code covers it.

You connect a client once, and every conversation in it can then reach your data. The phone app is a second client and asks for itself.

Connecting oneYour instance's address, then the sign-in you already have

The address is your backend's, with /mcp on the end. In claude.ai, open Settings, then Connectors, then Add custom connector, and paste it there.

Claude then sends your browser to a LifeOps sign-in page that asks for one username and one password. They are the ones you use for the web app. There is no second account and nothing to create: the page checks the same stored credential the browser login checks, with the same lockout behind it. Signing in sends you back to Claude, and the connector is live from the next message.

The dialog has no field for a token, which is why it works this way. A pasted token would be simpler and Claude's connector dialog does not accept one, so the sign-in page exists to hand out a token at the end of the handshake instead. The token it hands out is an ordinary LifeOps token, the same kind every other client uses.

That token does not expire. Disconnecting the connector in Claude stops that one client. Deleting the token on the backend stops every client at once, and your Claude can do that for you. Claude Code reaches the same door, configured on the machine it runs on rather than through this dialog.

What it reachesSeventeen tools, laid out by the page they answer about

Each tool answers about one part of the app, and the list is closed. A connected Claude cannot run a query of its own or reach a table no tool names.

• Day. One day at a time, with what you wrote on it and what was on your calendar, or a stretch of days for a question about a week. It reads any date you name and defaults to today in your own timezone.
• Journal. Search every day on record, by words in what you typed and in what a recording transcribed. Twenty entries come back at a time, and up to a hundred when more are asked for.
• Tasks. Everything still on your plate, meetings included, optionally narrowed to one life area. It also reads your status words, so it moves a task into a word you use rather than one that sounds right. Work you have already finished is not on that list, and the day it happened on has it instead.
• Goals. Every goal you have open, grouped by what your own status word means, with its life area, its kind, and the count of tasks under it that are open and done.
• Notes. Your notes by title, or searched, and then one note's full text. The ones you archived are left out until they are asked for, and a row that comes back from the shelf says on itself that it is off your list.
• Life areas. The areas you file things under, each with whatever you said it covers. This is what stops a write from inventing a category you never made.
• Events. The stretches of days you have named, grouped into what is underway, what is coming up and what you have not decided on. Each one comes back with its dates, whether its end is approximate, what you wrote on it, and the notes you filed under it. The ones already over and the ones you called off are left out until they are asked for.
• What has changed. Every recorded change over a window you name, in the order it happened. Two readings a week apart cannot tell you that a task went open, then stalled, then open again, and this does. Each change also names the hand that made it, so your own moves read differently from an agent's.
• A day that has already closed. Your tasks, your goals and your events as they stood at the end of a past day you name, replayed backward through that same record of changes, and one note's words as the trail of saves behind it holds them. An event you had not started yet is in that answer too: it was on your books that day, which is what the answer reports.

An everyday answer matches what the app puts in front of you. Goals you finished, life areas you retired and notes you archived are left out of it and come back when they are asked for, which is how the app's own screens behave.

A question about the past stops where the record stops. Ask about a day before your instance began recording changes and the answer refuses and gives you the date the record starts, rather than reconstructing weeks nobody wrote anything down for. Notes are stricter still. When a note's words changed after the day you asked about and no saved state from around then survives, the answer refuses rather than serving today's words under a past date.

What it writesOne write for each of your six stores, all of them signed

Six things it can write, one per store: a task, a goal, a document, an event, a journal entry, and a record of the conversation onto your day. That is the whole list. A task, a goal, a document or an event it names by its id is the one it changes; one it names without an id is a new one.

A task write sets any of the description, one of your statuses, a life area, a date and a note, and can take the task off one day’s list so it comes back tomorrow. That is everything you do to a task yourself apart from removing it. A goal write sets the name, the status, the area, the kind and the two dates. Rename a goal and every task, document and entry filed under the old name is refiled with it. A document write sets the title, the description, the area, the date and the words. An event write sets the name, whether you have committed to it, its two dates, whether that end is approximate, the area and the words on it; it cannot put one event inside another, which stays yours to arrange in the app.

A write only touches what it names. Moving a task to Done cannot blank its note, and renaming a document cannot empty it. Every field left out of a write is left exactly as it was, and there is no way for a connected Claude to empty a field at all. A document’s words and an event’s words are the two exceptions. Sending either replaces the whole body. A document’s previous words go into that document’s own history, as far back as that history reaches; an event’s do not, so read one before rewriting part of it. Dating a task is the one place a write moves a second thing: it moves the task’s status too, which is what dating something means here, and the answer says so.

Three things a write refuses rather than damages. A life area you do not use is refused. Everything here is filed by the area’s own identity, so a word with nothing behind it would read as filed on the page and unfiled to every query; a word you do use is stored in your spelling of it, whatever spelling the call arrived in. A date on a meeting mirrored from your calendar is refused too. That row’s date is the meeting’s start time, and a plain day would delete the hour. And a note holding a picture or a table refuses new words altogether, since the plain text a write sends has no room for either one and the picture would be gone. In each case the rest of that row is still writable, and the refusal names the call to make instead.

Everything it writes is marked as its own. A journal entry it writes is stamped as written by Claude and your journal shows it that way, next to the entries you typed. A conversation it logs is stamped as coming from a chat. A document it writes records that a Claude client wrote it. A task it adds arrives in a band of its own at the top of Tasks instead of joining your list at a rank nobody chose, and you keep it or discard it from there. Chat’s tasks arrive in the same band, since both write through the same tools. That band is still headed Added by your assistant on screen, the word the app used for Chat before the rename.

The same mark reaches the change record. A status a Claude moved is told apart from one you moved yourself, months later, without you having to remember which of you did it.

Deleting is not on the list, and neither is archiving. A connected Claude cannot remove a task, a goal, a document, an event or an entry, cannot put a document away, and cannot take one back off the shelf, so a wrong write is corrected by you on the page it landed on. Every write answers with what actually landed and what moved, for exactly that reason.

How answers are builtEvery answer explains itself before it gets read

A Claude reading your data has never seen your account. It does not know that Partial means you are still on it, that a goal with no tasks is ordinary, or that a list of goals left the finished ones out. So the answer states all of that itself.

• It says what it is and what it left out. A goal list says whether it is showing your open ones or all of them, and names the way to ask for the others. A filter that was applied is stated rather than left for a reader to notice.
• It says which moment it describes. Every answer states the instant it was read, timezone included. A question with no date in it is answered about now.
• It defines its own words. Every field in an answer is explained inside that answer, and so is every grouping. A count of zero says out loud that it is a real count and not a failed read.
• It uses your words and says so. Your status names, your life areas and your kinds come back with the data, marked as yours, so a chat says them back to you rather than translating them into something generic.
• A word it cannot place stays unplaced. A status your vocabulary does not define is grouped as standing unknown, with the instruction to ask you what it means. Guessing would put something you abandoned next to something you are working on.
• A refusal says what to do. A status that is not yours, or an id nothing has, comes back saying nothing was written, why, and the one call that fixes it.
• An unknown hand stays unknown. A change the record cannot attribute comes back marked with that word, and a Claude reading it is told plainly that the word does not mean you.

What it cannot reachAbsent on purpose, so absence is not a fault

These have no tool, and a connected Claude that says it cannot do them is right.

• Your photographs and videos. There is no tool that uploads one, reads one, or returns a transcript. Your media never leaves the app this way.
• Deleting anything, and archiving. A journal entry cannot be changed or removed once it exists, and nothing at all can be deleted: not a task, a goal, a note or an event. A note cannot be put on the shelf or taken off it either.
• Nesting an event inside another. It reads the nesting you arranged and cannot change it.
• Writing to your calendar. Adding an appointment or editing one has no tool. Reading is a different matter: a day's answer shows what was scheduled, and the change record shows an appointment that moved. Your calendar and your events are different things — an appointment is an hour, an event is a stretch of days.
• Settings, keys and your vocabulary. It reads your status words and life areas and cannot add, rename or remove one. Those are yours to author in the app.
• Device signals and sleep. What your phone and your Mac report is not exposed. Device signals covers where that data does go.
• Chat's history. The in-app Chat and a connected Claude share your data and share no conversation, and each is blind to the other's thread.