← Back to docs

Connect Claude to Vaulken

Claude supports the Model Context Protocol (MCP) in claude.ai, Claude Desktop and Claude Code. This guide shows how to connect each one to a Vaulken vault.

Prerequisites

  • A Vaulken account with at least one vault (create one here)
  • Your vault's MCP URL (found on the vault detail page in the dashboard)

Each vault has two MCP URLs:

  • Read-write (/mcp/vk_...): read, write, search and modify files
  • Read-only (/mcp/vk_...): read and search, no writes

Claude.ai

This is the quickest option because it needs no local setup.

  1. Open claude.ai
  2. Go to Settings > Integrations
  3. Click Add integration
  4. Paste your Vaulken MCP URL
  5. A browser popup asks you to authorize access. Click Authorize
  6. The vault is now available in all your Claude conversations

Use it for everyday conversations, quick lookups, and vault access from several devices.

Tips for Claude.ai

  • You can add several vaults, one integration each. Each one appears as a tool Claude can use in any conversation.
  • Ask Claude to "list files in my vault" to check that the connection works.

Claude Desktop

Setup

  1. Open your Claude Desktop configuration file:

    • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
    • Windows: %APPDATA%\Claude\claude_desktop_config.json
    • Linux: ~/.config/Claude/claude_desktop_config.json

  2. Add your vault to the mcpServers section:

{
  "mcpServers": {
    "my-vault": {
      "url": "https://vaulken.dev/mcp/vk_yourVaultId"
    }
  }
}
  1. Restart Claude Desktop
  2. On first use, Claude Desktop opens your browser for OAuth authorization
  3. Click Authorize on the Vaulken consent screen

Multiple vaults

Each vault gets its own entry and name:

{
  "mcpServers": {
    "work-notes": {
      "url": "https://vaulken.dev/mcp/vk_abc123"
    },
    "personal-vault": {
      "url": "https://vaulken.dev/mcp/vk_def456"
    }
  }
}

Tips for Claude Desktop

  • Use descriptive server names ("work-notes", "project-docs") so Claude knows which vault to use.
  • If authorization fails, delete the cached credentials and restart Claude Desktop. It reconnects on its own after a restart.

Claude Code

Setup

claude mcp add my-vault https://vaulken.dev/mcp/vk_yourVaultId

Claude Code runs the OAuth flow itself. On first use it opens your browser for authorization.

Verify the connection

claude mcp list

You should see your vault listed with a green status.

Usage examples

Once connected, you can mention the vault in any Claude Code session:

  • "Search my vault for files about authentication"
  • "Read the project roadmap from my vault"
  • "Save this implementation plan to my vault as docs/auth-plan.md"
  • "Find all markdown files modified this week"

Multiple vaults

claude mcp add work https://vaulken.dev/mcp/vk_abc123
claude mcp add notes https://vaulken.dev/mcp/vk_def456

Tips for Claude Code

  • Vault tools sit next to your local file tools, and Claude picks the right one from context.
  • Use read-only URLs for reference vaults you don't want modified.
  • Claude Code keeps MCP configurations across sessions.

Troubleshooting

"Authentication successful, but server reconnection failed"

This usually means the MCP server restarted during authorization. Restart your Claude client and try again.

Make sure you're logged in to your Vaulken dashboard in the same browser. The OAuth flow requires an active session.

Claude doesn't use vault tools

Be explicit: "Use the Vaulken vault to search for..." or "Read the file X from my vault". Claude picks up the habit after a few uses.

Token expired

Access tokens last one hour and are refreshed automatically. If you see token errors, restart your Claude client to trigger a fresh OAuth flow.

What Claude can do with your vault

Once connected, Claude has these tools:

Tool Description Scope
read Read a file's content, a markdown section or outline, or its metadata (PDFs via OCR, images via vision) Read
list List files with size and date, by path or most recent first Read
search_text Exact substring search across all text files and PDFs Read
search_semantic Meaning-based search using embeddings Read
get_related Find files similar to a given file Read
get_file_url Temporary download URL for large files (media, PDFs) Read
list_versions List previous versions of a file (requires S3 versioning on your bucket) Read
edit_file Write, append, replace a string, replace a markdown section, or set frontmatter Write
patch_file Apply several targeted edits to one file atomically Write
move_file Move or rename a file Write
delete_file Delete a file Write
restore_version Restore a previous version of a file (requires S3 versioning on your bucket) Write

Read tools are always available. Write tools require a read-write MCP URL. Versioning tools appear once S3 versioning is enabled on your bucket.

Next steps