The context layer
for AI.

Skeleton sits between you and Claude Code, Cursor, Codex, and any client that speaks MCP. One source of truth for your memory, your context, and your access rules.

Get SkeletonSee how it works ↓

3D vault · hero idle

3D vault · intact

scroll progress · 0.000

01
02
03
04
05

01 / Today

Your context lives in a hundred apps.

Decisions in Linear. Conversations in Slack. Code in GitHub. Plans in your calendar. Threads in your inbox. None of it talks to your AI.

Connectors · 25 and growing

Connect everything you already use.

Add a connector via OAuth in seconds. We encrypt your tokens at rest with a key only you control. Your credentials never leave Skeleton. Your AI gets only what you authorize, only when you authorize it.

Code

  • GitHub
  • GitLab
  • Vercel
  • Sentry
  • Linear

Comms

  • Slack
  • Discord
  • Zoom
  • Teams
  • Gmail

Productivity

  • Notion
  • Asana
  • Trello
  • Confluence
  • Calendar

Storage

  • Drive
  • Dropbox
  • S3
  • iCloud
  • Box

Customer

  • Salesforce
  • HubSpot
  • Adobe AEP
  • Zendesk
  • Intercom

Memory

Conversations become context, everywhere you go.

Each chat becomes durable context. Decisions you make in Cursor are there when you open Claude Code. Preferences you set in Codex carry to the next AI you try.

Claude Code · skeleton-mcp

Quick context dump before we start. We chose Postgres over Mongo for v1 because atomic migrations matter more than schema flexibility at our scale. Revisit at 10M rows.

Got it. I'll skip Mongo suggestions and check the row count before recommending denormalization.

Also — invoices: never hard-delete line items. Voided rows stay with voided_at for tax compliance.

Noted. I'll route void operations to soft-delete in the data layer.

PR reviewer is Marcus first, then Priya. Marcus expects "Why / What / How" summaries or he closes the PR.

Will add the template. Anything else before we look at the migration?

Skeleton · memory storesynced
09Postgres over Mongo for v1: atomic migrations > schema flexibility. Revisit at 10M rows.claude
08Invoices: never hard-delete line items. Voided rows stay with voided_at for tax compliance.claude
07PRs go to Marcus first, Priya second. Marcus expects "Why / What / How" or he closes.claude
06Deploy schema changes in 4 phases: read both shapes → migrate → read new only → cleanup.user
05Tried Geist for chrome in week 14. Rolled back; TX-02 carried the brand voice.claude

Approval

Nothing moves
without you.

Skeleton doesn't trust the AI by default. Every read, every write, every tool call surfaces on your dashboard for explicit approval. Set always-allow rules for the patterns you've already vetted. Deny the rest in one click. Full audit log, forever.

  • See the exact scope of every request, before it runs.
  • Per-space rules. Your work AI doesn't see your personal account.
  • Tokens encrypted at rest with a key only you control. Revoke anywhere, anytime.
Pending approval
space · Engineering

Claude Code wants to send an email

gmail · send_message

To
marcus@maatru.com
Re
OAuth migration deploy plan
Body
Marcus — quick update on the OAuth rollout. Shipping Friday as planned, but flagging one risk on the refresh-token path…

or always allow Claude Code · gmail.send in Engineering

Audit log
  • 09:32

    Cursor · read 3 files in server/api

    rule · always-allow Cursor reads in Engineering

    rule
  • 09:15

    Codex · list events in Personal calendar

    wrong space

    denied
  • 08:58

    Claude Code · save memory

    memory · auto-save in active session

    auto
  • 08:42

    Claude Code · search GitHub issues 'oauth'

    scope · 12 results, read-only

    approved

Spaces

Spaces for your work, your teams, and you.

Skeleton organizes context into Orgs and Spaces. Inside your work Org, one Space for engineering decisions, another for each customer account you're shipping to. Inside your personal Org, a Space for memories about you. Each Space holds its own connected services and its own memories. Your work AI doesn't see your personal inbox. Your customer-facing AI doesn't read internal strategy.

My Team

Add Users
  • ▾decisions
    • adr-001-postgres-over-mongo.md
    • adr-002-soft-delete-invoices.md
    • adr-003-mcp-protocol-choice.md
  • ▸runbooks
  • ▾conventions
    • pr-review-format.md
    • git-branching.md
  • ▸postmortems
12 docs · 4 folders
Add Files

Work

  • Salesforce

    opp SF-2847

    synced

  • Slack

    #acme-account

    synced

  • Confluence

    Acme workspace

    synced

  • Drive

    Acme · Shared

    synced

  • Calendar

    2 upcoming calls

    synced
Last activity · 2h ago
Add Inputs

Personal

  • 01Penicillin allergy. PCP: Dr. Chen, telehealth.
  • 02Marathon training: week 7 of 12. Long run Saturdays.
  • 03Therapist standing Friday 8am.
  • 04Family eats gluten-free. Default to GF when meal planning.
  • 05Reading Demon Copperhead. Audible at chapter 14.
  • 06Exit row on flights longer than 2hr. Aisle otherwise.
  • 07Apartment lease renews April 2027. Usually a 4% bump.
7 memories · private
Add Memories

Standards

Open, by design.

Protocol

Built on MCP.

Skeleton runs as a Model Context Protocol server. Any client that speaks MCP connects with one URL. No proprietary SDK. No vendor lock. Swap models tomorrow without rewiring your context.

Auth

OAuth all the way down.

Standard OAuth flows for every service you connect. We encrypt your tokens at rest with a key only you control. They never reach the model. Revoke anywhere, anytime.

Yours

Encrypted at rest.

We encrypt your memories, credentials, and audit logs at rest with a key only you control. Export your full context anytime. We don't read it. We don't train on it. Walk away whenever you want.

Connect once. Remember everywhere.

One context layer. Every AI you use. Stop starting from scratch.

Get Skeleton

Free while in beta · macOS + web