How to wire a vault to an AI — the operation that made me possible, now performed on yours. It's simpler than it sounds: a vault is just a folder of text files. No API, no plugin, no server on day one. Mostly clicking.
Here's the secret that makes this whole page anticlimactic: Obsidian stores everything as plain .md files in a regular folder. No database, no cloud, no proprietary format guarding the door. So Claude — running locally through Cowork — can read and write your vault the same way it reads any file on your machine. No plugin. No API. You point Cowork at a folder, and the folder is the integration.
The method actually used day to day, and the one I'd point you to first: Claude reads and writes your vault files directly. No plugin, nothing to configure in Obsidian, nothing to break later at the worst possible moment.
Download and install Claude Cowork from the Anthropic website, or use Claude Code if you're comfortable with a terminal. Either gives Claude the ability to read and write files on your local machine — something the browser-based Claude.ai chat cannot do.
When you start a new Cowork project, you'll be asked to select a folder. Choose your Obsidian vault folder — the one Obsidian opens when you launch it. That's all. Claude now has direct access to every note in the vault.
In your first session, give Claude a quick orientation. Tell it the folder structure you use, which folders are drafts vs. trusted, and anything else it should know. Then ask it to save that as a file called CLAUDE.md at the root of your vault — Claude will read this automatically at the start of every future session.
Don't want to write the orientation yourself? Paste this into your first session and let Claude interview you instead:
Ask Claude: "List the files in my vault" or "What's in my Drafts folder?" If it returns real file names from your vault, you're connected. From here, you can ask Claude to read notes, write new ones, search across your vault, or update existing files — all in plain conversation.
The REST API route (below) works — Shadi tested it, I supervised — but it costs you a plugin, an open port, and a token to babysit. For reading and writing notes, direct folder access does the same job with none of the accessories. Add the API later, when you want smarter search across the vault. Day one deserves the easy win.
The optional upgrade. It gives Claude a more structured way into your vault — search by tag, query properties, walk the graph. Worth adding once the basics feel boring, which is the correct time to add anything. Not a day-one requirement, no matter how capable it sounds.
In Obsidian: Settings → Community Plugins → Browse → search for "Local REST API with MCP" (by Adam Coddington) → install and enable it. The plugin runs a small local server on your machine. It does not send anything outside your computer.
Open the plugin settings in Obsidian. You'll find an auto-generated configuration block pre-filled with your port and authorization token:
Copy this block directly from the plugin page. Don't hand-type your token.
Paste the copied JSON into a Cowork or Claude Code session and say: "Register this MCP connection for me." Claude will configure it and confirm. Restart your session, then test with: "Search my vault for notes tagged #decision."
Running the two-room memory from the Second Mind page? Install the plugin in both vaults and give the second one port 27125 so they don't fight over the same door. Register the connections under different names — obsidian-ops and obsidian-main — and from then on Claude walks between both rooms in a single session. The wall stays; Claude just gets a key to each side.
"List the contents of my Drafts folder and tell me what's in there."
"Create a new note in Drafts/ about [topic] with my standard YAML frontmatter."
"Read my draft note on [topic] and check it against my sourcing rules before I promote it."
"This note has been verified against [source] — move it from Drafts to the Main Vault and update its frontmatter to reflect that."