← Back to docs

Connect Obsidian to Claude with S3 and Vaulken

This guide connects an Obsidian vault to an MCP-compatible AI tool (Claude, Cursor, Windsurf) through Vaulken and an S3 sync plugin. It takes about five minutes: sync the vault to a bucket, point Vaulken at the bucket, give the MCP URL to your AI.

What you'll need

  • An Obsidian vault
  • An S3-compatible bucket (AWS, Scaleway, MinIO, Cloudflare R2, Backblaze B2)
  • A Vaulken account (sign up free)

Step 1: Create a vault

Log in to your Vaulken dashboard and create a vault backed by your own S3 bucket:

  1. Click New vault
  2. Enter your S3 connection details: endpoint, region, bucket, access key, secret key
  3. Click Create vault

Step 2: Sync Obsidian to S3

The Remotely Save plugin syncs your Obsidian vault to S3-compatible storage.

Install the plugin

  1. In Obsidian, open Settings > Community Plugins > Browse
  2. Search for Remotely Save
  3. Install and enable

Configure S3

  1. Go to Remotely Save settings
  2. Select S3 or S3-compatible as the remote type
  3. Enable Path-style URL if your provider requires it (MinIO does)
  4. Fill in the same credentials you gave Vaulken:
    • Endpoint: the S3 endpoint (e.g., s3.fr-par.scw.cloud)
    • Region: the bucket region (e.g., fr-par)
    • Bucket: the bucket name
    • Access Key ID: your access key
    • Secret Access Key: your secret key
  5. Click Check connectivity
  6. Run the first sync

Your notes are now in the bucket.

Step 3: Connect your AI tool

Go to your vault's detail page in the Vaulken dashboard. You'll find two MCP URLs:

  • Read-write: your AI can read, search and edit your notes
  • Read-only: your AI can only read and search

Copy the URL you want and add it to your AI tool. The quickest way is claude.ai: go to Settings > Integrations > Add integration and paste the URL. Your notes are available in the next conversation.

For step-by-step setup instructions for Claude Desktop, Claude Code, and Cursor, see the Connect Claude to Vaulken guide.

If you have several vaults, name the MCP server obsidian-vault or similar so your AI knows which one to use.

What you can do now

Some things to ask once connected:

  • "Find all my notes about project X"
  • "Read my meeting notes from last week"
  • "Summarize what I know about topic Y across all my notes"
  • "Find notes related to this concept" (semantic search, matches on meaning)
  • "Write a summary of our conversation and save it to my vault" (read-write URL only)

PDFs in the vault are extracted with OCR, so your AI reads their text.

Tips

  • Remotely Save can sync on startup, on an interval, or by hand. A short interval (5 minutes) keeps the bucket close to your local vault.
  • Begin with the read-only MCP URL and switch to read-write once you trust your AI with edits.
  • The first read of a PDF takes a moment while OCR runs. Later reads come from the cache.

Going further

  • Set the vault type to Obsidian in the dashboard so your AI knows about frontmatter, wikilinks, canvas files and bases. See vault description and VAULKEN.md.
  • The set_frontmatter command of edit_file lets your AI manage Obsidian properties (tags, dates, status).
  • Enable S3 versioning on the bucket so any note your AI edits can be restored.