convez.ai
MCP Server
davidianstyle

openbrain-claude-starter

davidianstyle/openbrain-claude-starter

Ver no GitHub
InicialPluginScore · de 0 a 100

Portable template for a personal AI Chief of Staff built on an Obsidian LYT vault and Claude Code. One-script setup wires multi-account Gmail, Calendar, Drive, Slack, Asana, and Fathom MCPs, plus 21 Chief of Staff skills.

Instalação

$npx convez-ai davidianstyle/openbrain-claude-starter

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
1
Acompanhando
0
Saúde
42%
Último commit
há 1 mês
Versão
—
Issues abertas
13
Contribuidores
2

README

A tradução para o português deste README ainda está na fila.

OpenBrain

A personal AI Chief of Staff in an Obsidian vault, managed by Claude Code.

OpenBrain is a portable template for a Linking Your Thinking (LYT) knowledge base that doubles as an operating layer for Claude Code. Clone it, run one setup script, and you get:

  • A fully scaffolded Obsidian vault (Inbox, Spaces, Atlas, Sources, Templates)
  • 21 pre-built Chief of Staff skills (daily brief, inbox triage, capture meeting, etc.)
  • Multi-account MCP wiring for Gmail, Google Calendar, Google Meet, Google Drive/Docs/Sheets, Slack, Asana, and Fathom — any number of accounts per service
  • A people data model with cadence tracking, interaction logging, and alias resolution

Built for people who want Claude to act on their calendar, email, tasks, and notes the way a human chief of staff would — proactively, with context, and without constant re-briefing.


Prerequisites

  • macOS (Linux should work; untested)
  • git — usually xcode-select --install
  • Python 3.10+ — brew install python (system python also works)
  • Node.js 18+ — brew install node (or asdf / nvm)
  • Claude Code — installation instructions
  • Obsidian — download
  • Optional: GitHub CLI (brew install gh) if you want automated remote setup

Install

git clone https://github.com/davidianstyle/openbrain-claude-starter.git ~/OpenBrain
cd ~/OpenBrain
./bootstrap/setup.sh

The wizard will:

  1. Ask for your name and writing-voice blurb
  2. Customize CLAUDE.md with your details
  3. Create ~/.config/openbrain/ and copy launcher scripts
  4. Walk you through Google Cloud OAuth setup (one-time, 5 minutes)
  5. Loop through each service and ask which accounts to add:
    • "Add a Google account?" → y → paste email → browser OAuth → done
    • "Add another?" → repeat for as many as you want
    • Same for Slack workspaces, Asana, Fathom
  6. Register every MCP server with Claude Code
  7. Link the pre-commit hook
  8. Optionally enable auto git sync hooks (commit + push on stop, pull on start)
  9. Validate the install

Restart Claude Code and run /mcp to verify everything connected. Run /daily-brief as your first skill.


Adding accounts later

The wizard is not a one-shot. You can add services any time:

./bootstrap/lib/add-google-account.sh jane@newdomain.com
./bootstrap/lib/add-slack-workspace.sh newteam           # → newteam.slack.com
./bootstrap/lib/add-asana.sh personal                    # or work
./bootstrap/lib/add-fathom.sh
./bootstrap/lib/register-mcps.sh                         # re-sync ~/.claude.json

Each script is idempotent — safe to re-run.


What you get

Vault layout

~/OpenBrain/
├── + Inbox/                  # capture first, triage later
├── + Spaces/                 # MOCs (Maps of Content)
│   └── People.md             # people MOC (created on demand)
├── + Atlas/                  # atomic notes — the actual knowledge
│   ├── Daily/                # daily notes
│   ├── Weekly Reviews/       # weekly synthesis
│   ├── People/               # person notes
│   ├── Interactions/         # meeting/call/thread notes
│   ├── Ideas/
│   ├── Decisions/
│   ├── Goals/
│   ├── Places/
│   ├── Organizations/
│   └── Quotes/
├── + Sources/                # literature / reference notes
├── + Extras/
│   └── Templates/            # 14 note templates
├── + Archive/                # cold storage
├── CLAUDE.md                 # the operating manual Claude reads every session
├── Home.md                   # front door with MOC index
└── .claude/skills/           # 21 Chief of Staff skills

Skills

SkillPurpose
/daily-briefMorning briefing across all your calendars, mail, Slack, tasks
/daily-reviewEnd-of-day reconciliation — check off tasks, push Asana updates
/process-inboxTriage + Inbox/ + Gmail + Slack; auto-push tagged tasks to Asana
/meeting-prepAssemble a briefing for a meeting or 1:1
/capture-meetingTurn notes/transcript into an interaction note, update people
/capture-youtubeLiterature note from a YouTube video
/log-personCreate a person note, seeded from Gmail/Slack
/log-noteQuick-capture a thought as an atomic note
/log-interactionManual touchpoint log
/log-ideaRecord an idea
/log-decisionRecord a decision with context and alternatives
/log-goalCreate a goal with definition of done
/log-placeCreate a place note
/log-organizationCreate an organization note
/log-quoteSave a quote
/draft-follow-upDraft a reply email/Slack message (never sends)
/what-am-i-missingSurface overdue tasks, cadence misses, unanswered mail
/people-auditCadence health report + regenerate People MOC
/sync-peopleDiscovery pass across Gmail/Calendar/Slack to find unknown people
/weekly-reviewMonday synthesis
/push-openbrain-claude-starterGenericize vault improvements and open a PR against the template repo
/pull-openbrain-claude-starterPull latest template changes into the vault interactively
/asanaQuick view of upcoming Asana tasks with interactive check-off

Skills are markdown procedures — Claude reads the SKILL.md and performs the steps. No code execution.

Supported MCP servers

One stdio MCP server per (service × account) pair, so routing is explicit:

  • Google (google-mcp) — Gmail, Calendar, Meet, Drive, Docs, Sheets, Slides — one consolidated server per Google account
  • Slack (slack-mcp) — one per workspace
  • Asana (asana-mcp) — personal + work
  • Fathom (fathom-mcp) — single instance

All launched via ~/.config/openbrain/lib/*-mcp.sh wrappers that source ~/.config/openbrain/.env.


Design principles

  • Capture first, organize later. Everything starts in + Inbox/.
  • Atomic notes. One idea per note. If it wants to split, split it.
  • Links over folders. Structure comes from [[wikilinks]] and MOCs.
  • Never delete, always archive. Move to + Archive/, never rm.
  • Git is the sync layer. No Obsidian Sync. Auto git sync (commit + push on stop, pull on start) is available as an opt-in during setup. Set OPENBRAIN_AUTOPUSH=0 to keep auto-commits local-only (no push).
  • Skills are markdown procedures. Claude reads them and performs the steps.
  • People are first-class entities. Every person gets a note. Interactions link back. Cadence is tracked.
  • Multi-account by default. Every external service is wired per-account with routing tags.

Troubleshooting

See bootstrap/README.md for:

  • Re-running parts of the wizard
  • Google OAuth gotchas (admin-managed Workspace accounts, "unverified app" screens)
  • Slack workspace admin approval
  • Rotating tokens
  • Removing an account

Credits

Developed by @davidianstyle as the portable template extracted from his personal OpenBrain vault.

The underlying LYT methodology is from Nick Milo.

Ver o original no GitHub

Bom para

Relacionados