Obsidian for Claude Code: How to Use Your Vault as AI Memory
Claude Code reads files. Obsidian stores everything as files. Put those two facts together and...
Claude Code reads files. Obsidian stores everything as files. Put those two facts together and you get one of the most popular AI setups of 2026: your Obsidian vault as a memory layer that Claude Code can read, search, and write back to. Here's how the setup works, what it's good at, and where it stops.
Why does Obsidian work so well with Claude Code?
Because they share a foundation: plain files on your disk. Claude Code is a terminal agent that operates on whatever folder you open it in — reading files, writing them, running commands, maintaining context across a long task. An Obsidian vault is exactly that: a folder of Markdown files. No format conversion, no API, no upload. Claude Code can read your notes the same way it reads source code.
Two Claude Code features make it especially good for this. It reads a CLAUDE.md file at the start of every session automatically, so you can give it standing instructions about your vault. And it has full file-system access in the directory you open, with persistent context across a working session — so it can genuinely work in your notes, not just answer questions about them.
How do you set it up?
Three layers, simplest first:
1. Direct vault access. Open your vault folder in Claude Code (claude from the vault directory). It can now read and write your notes. Add a CLAUDE.md at the root describing your vault: the folder structure, what each area is for, and what to ignore (templates, attachments, archives — use a .claudeignore to keep the agent from wasting tokens on them). This alone gets you a working setup.
2. MCP bridge. Install an Obsidian MCP server or the Local REST API plugin and register it with Claude Code. This gives the agent dedicated vault tools — structured search, read, create, update — rather than raw file reads, and it works even when Obsidian itself is closed. Worth it once your vault grows past a few hundred notes.
3. Vault structure for AI. Use consistent frontmatter (note type, tags, project, status, date) so Claude Code can filter and find notes without reading every file. A well-structured vault is the difference between an agent that navigates your knowledge and one that flails through it.
What can you actually do with it?
The useful workflows people build:
- Daily notes: "Create today's daily note from my template and carry over unfinished tasks from yesterday."
- Capture without leaving flow: "Add 'compare backup strategies' to my Inbox."
- Meeting extraction: "Read my note from the 2026-04-08 team meeting and pull out action items assigned to me."
- Weekly review: "Read this week's daily notes, summarize what got done and what didn't, and draft next week's priorities."
The shift people describe is from a notebook that stores what you wrote to an assistant that finds it, connects it, and acts on it. For knowledge work in the terminal, it's a genuine step up.
Where does it stop?
At the edges of Claude Code and your local machine.
It's tied to Claude Code. The smoothness comes from Claude Code's direct file access. Your Claude.ai browser session, ChatGPT, or another tool doesn't get this — they can't read your local vault without separate bridges, and some can't at all. So the context that makes Claude Code so capable doesn't follow you to your other tools.
It's still manual. The vault reflects what you write and maintain. Claude Code can help you maintain it, but the structure, the upkeep, and the discipline are yours. And the most dynamic context — your current situation — is the hardest to keep current in a file-based vault.
It's about your knowledge, not your identity. This setup is brilliant for "work with my notes." It's not designed to give every AI tool a current understanding of who you are and what you're working on — that's a different layer.
Obsidian + Claude Code, plus a context layer
For terminal knowledge work, Obsidian + Claude Code is hard to beat — set it up, structure your vault, enjoy it. But if you also want your other tools (Claude.ai, ChatGPT, Cursor) to understand you without rebuilding this rig for each one, pair it with a context layer that serves your identity and current context to any tool through MCP. Obsidian holds your knowledge for Claude Code; the context layer carries who you are everywhere else. Different jobs, and they stack cleanly.
→ The non-terminal version: Obsidian for Claude
→ Optimizing what Claude Code loads: How to Optimize Context in Claude Code
→ Carry your context across every tool, not just Claude Code — Unabyss →
