A minimal Obsidian vault wired for Claude. Folder structure, frontmatter contract, and three working skills — drop-in for any existing vault.
This is not a preview. Clone it, point Claude at it, and you have a working AI-assisted knowledge base in ten minutes.
Most Obsidian + AI setups use a plugin that pipes selected text into an LLM. That works, but it caps the AI at whatever a single note can hold.
This vault takes a different route: Claude reads the vault as a filesystem. It can list folders, read notes, follow wiki-links, write new files, and update frontmatter. The vault becomes the agent's working memory.
The structure here is opinionated on purpose. Numeric prefixes give Claude a stable map. The frontmatter contract gives it queryable metadata. The three skills give it ready-made loops.
00-Inbox/ Capture zone. New notes land here.
10-Projects/ Active work with a defined outcome.
20-Areas/ Ongoing responsibilities, no end date.
30-Resources/ Reference material, evergreen notes.
40-Archive/ Finished or dormant.
90-System/ Templates, handbook, contracts.
.claude/skills/ Three Claude skills (see below).
.mcp.json Filesystem MCP config for Claude Desktop.
CLAUDE.md Instructions Claude reads on every session.
Contents:
- Example notes across all folders (one daily, one weekly, one project, one area, three resources, plus inbox samples)
- 4 templates (daily, weekly, project, evergreen)
- 3 skills:
capture,daily-review,weekly-digest - Frontmatter contract:
type,status,created,updated,tags(plusmaturityfor evergreen notes) - 5-page handbook covering setup, the loop, and the first week
- Obsidian (free) — obsidian.md
- Claude Code (recommended) or Claude Desktop — any Claude plan that includes one of them
That's it. No paid plugins. No extra accounts to create.
1. Clone the repo
git clone https://github.com/SecondBrainProd/second-brain-starter.git my-vault
cd my-vault2. Open as an Obsidian vault
Obsidian → Open folder as vault → select my-vault.
3. Connect Claude
Claude Code — nothing to install. Claude Code reads and writes files out of the box. Launch it from the vault folder and you're done:
cd my-vault
claudeIt picks up CLAUDE.md and the skills automatically.
Claude Desktop — one config step. Claude Desktop needs an MCP server to see your files. This repo ships a ready config in .mcp.json that uses the official filesystem server (@modelcontextprotocol/server-filesystem). Copy the server entry into your claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json), replace /absolute/path/to/your/vault with your actual vault path, and restart Claude Desktop.
4. Verify
Ask Claude: “What folders are in my vault?”
If it lists 00-Inbox, 10-Projects, and the rest — you're done.
Capture anything. In Claude: “Capture this conversation as a note.” The capture skill writes a properly formatted note into 00-Inbox/ with frontmatter filled in.
Daily review. End of day: “Run my daily review.” The skill scans today's inbox, links new notes to existing ones, and drafts tomorrow's first task.
Weekly digest. End of week: “Give me my weekly digest.” The skill walks active projects, surfaces stalled work, and compiles what shipped.
The full workflow is in 90-System/Handbook.md. Read it once before you start.
The vault is plain files on your disk. Nothing syncs, uploads, or indexes in the background — this setup has no server and no account to create beyond the Claude plan you already use.
The one tradeoff worth understanding: when Claude reads a note during a conversation, that note's text goes to Anthropic's API as part of the conversation — the same as pasting it into a chat. Claude reads files only when you ask it to do something, so notes it never touches don't leave your machine. If some notes must stay local no matter what, keep them in a folder you exclude in CLAUDE.md.
That's the honest version. It isn't “100% private” — it's private until you ask Claude to read something, and you control what that is.
Already have a vault? Copy these into yours and you're set:
.claude/skills/→ adds the three skills90-System/Handbook.md→ reference90-System/Templates/→ templates (rename to match your existing folders).mcp.json→ only if you use Claude Desktop (edit the path)
The numeric prefixes are a recommendation, not a requirement. The skills work against any folder names if you update the paths in CLAUDE.md.
This is a skeleton — clean, working, and empty by design. It gives you the architecture and the three loops, not a lived-in system. Filling it with your own notes is the whole point, and that part is yours.
Built and maintained by Second Brain Production. A paid, fully built-out version of the same system lives at secondbrain.tools — worth a look only if you want to see the architecture at real scale. This starter stands on its own.
- Code (skills, MCP config, scripts) — MIT, see
LICENSE - Content (example notes, handbook, templates) — CC0, see
LICENSE-CONTENT
You can fork, ship, sell derivatives. Attribution appreciated but not required.
Issues and PRs welcome. Keep changes minimal — this is a starter, not a framework.