Documentation

Skeleton is the utility that carries your context. You connect your sources once, and the memory, documents, and approvals you build up follow you into every AI tool you use — instead of starting each assistant from zero.

This page covers the concepts and the setup. For the exact tools an AI client can call, see the tool catalog.

01. What Skeleton is

An AI assistant is only as useful as what it knows about you. Today that knowledge is trapped inside whichever product you happened to type it into. Skeleton is a context substrate that sits underneath your AI tools: you own the context, and you decide which assistant can read which part of it.

Three things make it work:

• One place your context lives. Memories, documents, people, and files you bring in — organised by you, versioned, and attributable.
• A standard way in. Skeleton speaks the Model Context Protocol (MCP), so any MCP client — Claude, Claude Code, Cursor, ChatGPT, VS Code, Windsurf, and others — connects to the same context.
• An approval gate in front of everything sensitive. Nothing syncs on its own. An assistant asking for something you haven't already allowed waits for you to approve it.

02. Core concepts

Space

A space is a container for context — one for personal life, one for a company, one per project. Memories and documents belong to a space, so work context never leaks into a personal conversation. Your identity is global and travels with you; everything else is scoped to a space. Every session has one active space, and you can switch it at any time.

Memory

A memory is a durable fact you want carried forward — a preference, a constraint, a decision, something about how you work. Memories are stored verbatim with version history, so you can always see what changed and when. Every memory write is attributed to whoever made it, whether that was you or an assistant acting on your behalf.

Document

Documents are the longer-form context in a space — notes, briefs, plans, reference material, uploaded files. They're editable in the dashboard, searchable, and readable by any AI client you've connected and approved.

Connector

A connector is a source Skeleton can reach on your behalf: Gmail, Google Calendar, GitHub, Jira, Figma, a database, or an on-device reader for things that live on your own Mac. Connecting one doesn't copy your data — it opens a route Skeleton can use when you approve a request.

Approval

When an assistant asks for something gated, Skeleton raises an access request. It appears live in the dashboard and on your phone; you approve or deny, and the assistant proceeds or doesn't. You can grant a standing allowance for calls you approve routinely, and revoke it later.

03. Getting started

• Create your account. Sign in at skeleton.mother.tech. Everything starts in your Personal space.
• Tell it who you are. Your profile — name, what you do, how you want assistants to talk to you — is the identity every connected client reads first.
• Add a space per context.Keep work, personal, and per-client context in separate spaces from the start; it's much harder to untangle later.
• Connect a source or two. Start with the ones you live in — mail, calendar, code.
• Pair your AI client and ask it what it knows about you.

04. Connecting an AI client

Skeleton exposes a single MCP endpoint. Point any MCP client at it and authenticate in the browser — there are no tokens to copy into config files, and no API key to leak:

https://api.skeleton.mother.tech/mcp

For Claude Code, that's one command:

claude mcp add --transport http skeleton https://api.skeleton.mother.tech/mcp

For Claude Desktop, Cursor, Windsurf, ChatGPT, and Lovable, paste the same URL into the app's connector or MCP settings. VS Code uses its own config shape. The Connections page has copy-ready instructions per client, and shows every client currently paired to your account so you can revoke one at any time.

The first call from a new client opens a browser window asking you to authorise it and choose which spaces it can reach. A client only ever sees the tools it's entitled to.

05. Connectors

Connectors are how Skeleton reaches the systems you already use. Browse and connect them from the Connections page. Each one carries its own controls:

• Read and write are separate. A connector can be allowed to read without ever being allowed to write, and the toggle is enforced before any request goes out.
• Status is verified, not assumed. A connector reads as live only when Skeleton has actually reached it.
• Credentials are encrypted at rest and can be revoked from the same page. Revoking removes the route immediately.
• Some sources stay on your machine. On-device connectors — your messages, photos, notes, and local files — are read by the desktop app on your Mac. Local archives are parsed on-device; the raw files never leave it.

06. Approvals and consent

Skeleton never syncs on its own. Data moves when you approve a specific request, or when you've set a standing rule that covers it.

• Identity reads need no approval — your name, your spaces, your stated preferences are things you authored for assistants to read.
• Memory and connector calls are gated. The request shows you which client is asking, for what, and in which space.
• Approve on any device.Requests broadcast live to the dashboard and the mobile app, so you're not stuck at your desk.
• Everything is logged. Every gated call leaves an audit record — what was asked, what was decided, and what was returned.

07. Collections

A collection is a table of records built from context you already have — people, companies, tasks, receipts, whatever you define. Turn one on for a space and Skeleton fills it from the records in that space. Nothing is created behind your back: proposed rows arrive marked as proposals for you to accept or reject.

08. Sharing and workspaces

A workspace groups spaces for an organisation or a team. Spaces can be shared with other people, and a shared space carries its context to every member — their assistants read the same documents and collections you do, under the same approval rules.

Sharing is always explicit. A space you own stays yours until you invite someone into it, and access can be withdrawn.

09. Where Skeleton runs

• Web. The full dashboard at skeleton.mother.tech — spaces, memories, documents, connections, approvals.
• Desktop (macOS). The same dashboard in a native app, plus the on-device readers that let Skeleton reach local sources without uploading them.
• Mobile (iOS and Android). A companion app for approving requests and checking in on your spaces while away from a computer.
• Any MCP client. The point of the whole thing — your context, in the assistant you already use.

10. Your data

• It stays yours. Skeleton holds context on your behalf so you can lend it to an assistant — you can edit or delete any of it.
• Read-only at the boundary. Connected sources are read unless you explicitly grant write access to one.
• Nothing moves without a decision. No background sync, no silent copying between spaces.
• History is kept. Memories and documents are versioned, so an edit is recoverable and a deletion is auditable.

Full detail on collection, processing, and retention is in the Privacy Policy; the terms of service are here.

11. Getting help

Questions, problems, or something that looks wrong — skeleton.support@mother.tech. For pricing, teams, and enterprise deployments, skeleton.sales@mother.tech.