MCPVault

A universal AI bridge for Obsidian vaults using the Model Context Protocol (MCP) standard. Connect any MCP-compatible AI assistant to your knowledge base - works with Claude, ChatGPT, and future AI tools. This server provides safe read/write access to your notes while preventing YAML frontmatter corruption.

Universal Compatibility

Works with any MCP-compatible AI assistant including Claude Desktop, Claude Code, ChatGPT Desktop (Enterprise+), OpenCode, Gemini CLI, OpenAI Codex, IntelliJ IDEA 2025.1+, Cursor IDE, Windsurf IDE, and future AI platforms that adopt the MCP standard.

https://github.com/user-attachments/assets/657ac4c6-1cd2-4cc3-829f-fd095a32f71c

Quick Start (5 minutes)

1. Install Node.js runtime:

   # Download from https://nodejs.org (v18.0.0 or later)
   # or use a package manager like nvm, brew, apt, etc.
   

2. Test the server:

   If using the published package:

   npx @modelcontextprotocol/inspector npx @bitbonsai/mcpvault@latest /path/to/your/vault
   

3. Configure your AI client:

   Claude Desktop - Copy this to claude_desktop_config.json:

   {
     "mcpServers": {
       "obsidian": {
         "command": "npx",
         "args": ["@bitbonsai/mcpvault@latest", "/path/to/your/vault"]
       }
     }
   }
   

   Claude Code - Copy this to ~/.claude.json:

   {
     "mcpServers": {
       "obsidian": {
         "command": "npx",
         "args": ["@bitbonsai/mcpvault@latest", "/path/to/your/vault"],
         "env": {}
       }
     }
   }
   

   OpenCode - Copy this to ~/.config/opencode/opencode.json

   {
     "mcp": {
       "obsidian": {
         "type": "local",
         "command": [
           "npx",
           "@bitbonsai/mcpvault@latest",
           "/path/to/your/vault/"
         ],
         "enabled": true
       }
     }
   }
   

   Replace /path/to/your/vault with your actual Obsidian vault path.

   For other platforms, see detailed configuration guides below.

4. Test with your AI:

   • "List files in my Obsidian vault"
   • "Read my note called 'project-ideas.md'"
   • "Create a new note with today's date"

Success indicators: Your AI should be able to list files and read notes from your vault.

Why MCPVault?

Universal AI Compatibility

Built on the open Model Context Protocol standard, MCPVault is not locked to any single AI provider. As more AI assistants adopt MCP, your investment in this tool grows more valuable. Today it works with Claude and ChatGPT - tomorrow it will work with whatever AI tools emerge.

Future-Proof Your Knowledge Base

Instead of waiting for each AI company to build Obsidian integrations, MCPVault provides a universal adapter that works with any MCP-compatible assistant. One tool, endless possibilities.

Open Standard, No Lock-in

MCP is an open protocol. You're not tied to any specific vendor or platform. Your notes remain yours, accessible through any compatible AI assistant.

Features

• ✅ Safe frontmatter parsing and validation using gray-matter with AST-aware updates that preserve raw formatting for unmodified fields
• ✅ Path filtering to exclude .obsidian directory and other system files
• ✅ Complete MCP toolkit: 14 methods covering all vault operations
  • File operations: read_note, write_note, patch_note, delete_note, move_note, move_file
  • Directory operations: list_directory
  • Batch operations: read_multiple_notes
  • Search: search_notes with multi-word matching and BM25 relevance reranking
  • Metadata: get_frontmatter, update_frontmatter, get_notes_info, get_vault_stats
  • Tag management: manage_tags (add, remove, list)
• ✅ Write modes: overwrite, append, prepend for flexible content editing
• ✅ Tag management: add, remove, and list tags in notes
• ✅ Safe deletion with confirmation requirement to prevent accidents
• ✅ Automatic path trimming to handle whitespace in inputs
• ✅ TypeScript support with Node.js runtime (using tsx for execution)
• ✅ Comprehensive error handling and validation
• ✅ Token-optimized responses: 40-60% smaller responses with minified field names and compact JSON (v0.6.3+)
• ✅ Optional pretty-printing: Set prettyPrint: true for human-readable debugging
• ✅ Performance optimized: No unnecessary token consumption, efficient for large vaults
• ✅ Zero dependencies: No Obsidian plugins required, works with any vault structure

Prerequisites

• Node.js runtime (v18.0.0 or later)
• An Obsidian vault (local directory with .md, .markdown, .txt, .base, or .canvas files)
• MCP-compatible AI client (Claude Desktop, ChatGPT Desktop, Claude Code, etc.)

Installation

For End Users (Recommended)

No installation needed! Use npx to run directly:

npx @bitbonsai/mcpvault@latest /path/to/your/obsidian/vault

If you omit the vault path, the server uses your current working directory as the vault root.

For Developers

1. Clone this repository
2. Use the correct Node.js version:

nvm use  # Uses Node 24 from .nvmrc

3. Install dependencies with npm:

npm install  # Corepack automatically uses npm 10.9.0

4. Test locally with MCP inspector:

npx @modelcontextprotocol/inspector npm start /path/to/your/vault

Pro tip: Use MCP Inspector to test all server functionality before configuring with AI clients:

# Install globally for easier access
npm install -g @modelcontextprotocol/inspector

# Test with any vault
mcp-inspector npx @bitbonsai/mcpvault@latest /path/to/your/vault

Usage

Running the Server

End users:

npx @bitbonsai/mcpvault@latest
npx @bitbonsai/mcpvault@latest /path/to/your/obsidian/vault
npx @bitbonsai/mcpvault@latest ./Vault

Developers:

npm start
npm start /path/to/your/obsidian/vault
npm start ./Vault

AI Client Configuration

Claude Desktop

Add to your Claude Desktop configuration file:

Single Vault:

{
  "mcpServers": {
    "obsidian": {
      "command": "npx",
      "args": [
        "@bitbonsai/mcpvault@latest",
        "/Users/yourname/Documents/MyVault"
      ]
    }
  }
}

Multiple Vaults:

{
  "mcpServers": {
    "obsidian-personal": {
      "command": "npx",
      "args": [
        "@bitbonsai/mcpvault@latest",
        "/Users/yourname/Documents/PersonalVault"
      ]
    },
    "obsidian-work": {
      "command": "npx",
      "args": [
        "@bitbonsai/mcpvault@latest",
        "/Users/yourname/Documents/WorkVault"
      ]
    }
  }
}

Configuration File Locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: C:\Users\{username}\AppData\Roaming\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

You can also access this through Claude Desktop → Settings → Developer → Edit Config

ChatGPT Desktop

Requirements: ChatGPT Enterprise, Education, or Team subscription (not available for individual Plus users)

ChatGPT uses MCP through Deep Research and developer mode. Configuration is done through the ChatGPT interface:

1. Access ChatGPT developer mode (beta feature)
2. Configure MCP servers through the built-in MCP client
3. Create custom connectors for your organization

Note: ChatGPT Desktop's MCP integration is currently limited to enterprise subscriptions and uses a different setup process than file-based configuration.

Claude Code

Claude Code uses .claude.json configuration file:

User-scoped (recommended): Edit ~/.claude.json:

{
  "mcpServers": {
    "obsidian": {
      "command": "npx",
      "args": ["@bitbonsai/mcpvault@latest", "/path/to/your/vault"],
      "env": {}
    }
  }
}

Project-scoped: Edit .claude.json in your project or add to the projects section:

{
  "projects": {
    "/path/to/your/project": {
      "mcpServers": {
        "obsidian": {
          "command": "npx",
          "args": ["@bitbonsai/mcpvault@latest", "/path/to/your/vault"]
        }
      }
    }
  }
}

Using Claude Code CLI:

claude mcp add obsidian --scope user npx @bitbonsai/mcpvault /path/to/your/vault

Goose Desktop

On Goose Desktop settings, click Add custom extension, and on the command field add:

npx @bitbonsai/mcpvault@latest /path/to/your/vault

Other MCP-Compatible Clients (2025)

Confirmed MCP Support:

• IntelliJ IDEA 2025.1+ - Native MCP client support
• Cursor IDE - Built-in MCP compatibility
• Windsurf IDE - Full MCP integration
• Zed, Replit, Codeium, Sourcegraph - In development
• Microsoft Copilot Studio - Native MCP support with one-click server connections

Most modern MCP clients use similar JSON configuration patterns. Refer to your specific client's documentation for exact setup instructions.

Examples

Ask your AI assistant about your notes:

• "What files are in my Obsidian vault?"
• "Read my note called 'project-ideas.md'"
• "Show me all notes with 'AI' in the title"

Have your AI assistant help with note management:

• "Create a new note called 'meeting-notes.md' with today's date in the frontmatter"
• "Append today's journal entry to my daily note"
• "Prepend an urgent task to my todo list"
• "Add the tags 'project' and 'urgent' to my task note"
• "List all tags in my research note"
• "Remove the 'draft' tag from my completed article"
• "List all markdown files in my 'Projects' folder"
• "Delete the old draft note 'draft-ideas.md' (with confirmation)"

Advanced Use Cases:

• Knowledge Synthesis: "Summarize all my research notes tagged with 'machine-learning' from the last month"
• Project Management: "Update the status in all project notes to 'completed' and add today's date"
• Content Analysis: "Find all notes that mention 'API design' and create a comprehensive guide"
• Smart Tagging: "Review my untagged notes and suggest appropriate tags based on content"

Troubleshooting

Common Issues

"command not found: npx"

• Solution: Install Node.js runtime from nodejs.org
• Alternative: Use global install: npm install -g @bitbonsai/mcpvault

"File not found" when paths look correct

• Cause: The server is using the wrong vault root
• Solution: Either run the command from your vault directory or pass the vault path explicitly

"Permission denied" errors

• Cause: Insufficient file system permissions
• Solution: Ensure the vault directory is readable/writable by your user

"Path traversal not allowed"

• Cause: Trying to access files outside the vault
• Solution: All file paths must be relative to the vault root

AI client not recognizing the server

1. Check the configuration file path is correct for your OS
2. Ensure JSON syntax is valid (use a JSON validator)
3. Restart your AI client after configuration changes
4. Check your AI client's logs for error messages
5. Verify your AI client supports MCP (Model Context Protocol)

".obsidian files still showing up"

• Expected: The path filter automatically excludes .obsidian/** patterns
• If still seeing them: The filter is working as designed for security

Debug Mode

Run with error logging:

npx @bitbonsai/mcpvault /path/to/vault 2>debug.log

Getting Help

• Open an issue on GitHub
• Include your OS, Node.js version, and error messages
• Provide the vault directory structure (without sensitive content)

Testing

Run the test suite:

npm test

API Methods

read_note

Read a note from the vault with parsed frontmatter.

Request:

{
  "name": "read_note",
  "arguments": {
    "path": "project-ideas.md",
    "prettyPrint": false
  }
}

Response (optimized for tokens):

{
  "fm": {
    "title": "Project Ideas",
    "tags": ["projects", "brainstorming"],
    "created": "2023-01-15T10:30:00.000Z"
  },
  "content": "# Project Ideas\n\n## AI Tools\n- MCP server for Obsidian\n- Voice note transcription\n\n## Web Apps\n- Task management system"
}

Response (with prettyPrint: true):

{
  "fm": {
    "title": "Project Ideas",
    "tags": ["projects", "brainstorming"],
    "created": "2023-01-15T10:30:00.000Z"
  },
  "content": "# Project Ideas\n\n## AI Tools\n- MCP server for Obsidian\n- Voice note transcription\n\n## Web Apps\n- Task management system"
}

write_note

Write a note to the vault with optional frontmatter and write mode.

Write Modes:

• overwrite (default): Replace entire file content
• append: Add content to the end of existing file
• prepend: Add content to the beginning of existing file

Request (Overwrite):

{
  "name": "write_note",
  "arguments": {
    "path": "meeting-notes.md",
    "content": "# Team Meeting\n\n## Agenda\n- Project updates\n- Next milestones",
    "frontmatter": {
      "title": "Team Meeting Notes",
      "date": "2023-12-01",
      "tags": ["meetings", "team"]
    },
    "mode": "overwrite"
  }
}

Request (Append):

{
  "name": "write_note",
  "arguments": {
    "path": "daily-log.md",
    "content": "\n\n## 3:00 PM Update\n- Completed project review\n- Started new feature",
    "mode": "append"
  }
}

Response:

{
  "message": "Successfully wrote note: meeting-notes.md (mode: overwrite)"
}

patch_note

Efficiently replace an exact string inside an existing note without rewriting the full file.

Request:

{
  "name": "patch_note",
  "arguments": {
    "path": "meeting-notes.md",
    "oldString": "- Next milestones",
    "newString": "- Next milestones (owner: Alex)",
    "replaceAll": false
  }
}

Response (success):

{
  "success": true,
  "path": "meeting-notes.md",
  "message": "Successfully replaced 1 occurrence",
  "matchCount": 1
}

Response (multiple matches with replaceAll=false):

{
  "success": false,
  "path": "meeting-notes.md",
  "message": "Found 3 occurrences of the string. Use replaceAll=true to replace all occurrences, or provide a more specific string to match exactly one occurrence.",
  "matchCount": 3
}

list_directory

List files and directories in the vault.

Note: this includes non-note filenames (for example pdf, png, jpg) so AI assistants can see vault structure, but note tools like read_note and write_note still operate on note files only (.md, .markdown, .txt, .base, .canvas).

Request:

{
  "name": "list_directory",
  "arguments": {
    "path": "Projects",
    "prettyPrint": false
  }
}

Response (optimized):

{
  "dirs": ["AI-Tools", "Web-Development"],
  "files": ["project-template.md", "roadmap.md"]
}

delete_note

Delete a note from the vault (requires confirmation for safety).

Request:

{
  "name": "delete_note",
  "arguments": {
    "path": "old-draft.md",
    "confirmPath": "old-draft.md",
    "trashMode": "local"
  }
}

Response (Success):

{
  "success": true,
  "path": "old-draft.md",
  "message": "Successfully moved note to vault trash: old-draft.md"
}

Trash modes:

• none (default): permanent delete
• local: move to .trash inside the vault, preserving folder structure
• system: move to the OS trash/recycle bin

Response (Confirmation Failed):

{
  "success": false,
  "path": "old-draft.md",
  "message": "Deletion cancelled: confirmation path does not match. For safety, both 'path' and 'confirmPath' must be identical."
}

⚠️ Safety Note: The confirmPath parameter must exactly match the path parameter to proceed with deletion. This prevents accidental deletions.

get_frontmatter

Extract only the frontmatter from a note without reading the full content.

Request:

{
  "name": "get_frontmatter",
  "arguments": {
    "path": "project-ideas.md",
    "prettyPrint": false
  }
}

Response (optimized, returns frontmatter directly):

{
  "title": "Project Ideas",
  "tags": ["projects", "brainstorming"],
  "created": "2023-01-15T10:30:00.000Z"
}

manage_tags

Add, remove, or list tags in a note. Tags are managed in the frontmatter and inline tags are detected.

Request (List Tags):

{
  "name": "manage_tags",
  "arguments": {
    "path": "research-notes.md",
    "operation": "list"
  }
}

Request (Add Tags):

{
  "name": "manage_tags",
  "arguments": {
    "path": "research-notes.md",
    "operation": "add",
    "tags": ["machine-learning", "ai", "important"]
  }
}

Request (Remove Tags):

{
  "name": "manage_tags",
  "arguments": {
    "path": "research-notes.md",
    "operation": "remove",
    "tags": ["draft", "temporary"]
  }
}

Response:

{
  "path": "research-notes.md",
  "operation": "add",
  "tags": ["research", "ai", "machine-learning", "important"],
  "success": true,
  "message": "Successfully added tags"
}

search_notes

Search for notes in the vault by content or frontmatter with multi-word matching and BM25 relevance reranking.

Request:

{
  "name": "search_notes",
  "arguments": {
    "query": "machine learning",
    "limit": 5,
    "searchContent": true,
    "searchFrontmatter": false,
    "caseSensitive": false,
    "prettyPrint": false
  }
}

Response (optimized with minified field names):

[
  {
    "p": "ai-research.md",
    "t": "AI Research Notes",
    "ex": "...machine learning...",
    "mc": 2,
    "ln": 15,
    "uri": "obsidian://open?vault=MyVault&file=ai-research.md"
  }
]

Field names:

• p = path
• t = title
• ex = excerpt (21 chars context)
• mc = match count
• ln = line number
• uri = Obsidian deep link for quick opening

move_note

Move or rename a note in the vault (.md, .markdown, .txt, .base, .canvas).

Request:

{
  "name": "move_note",
  "arguments": {
    "oldPath": "drafts/article.md",
    "newPath": "published/article.md",
    "overwrite": false
  }
}

Response:

{
  "success": true,
  "oldPath": "drafts/article.md",
  "newPath": "published/article.md",
  "message": "Successfully moved note from drafts/article.md to published/article.md"
}

move_file

Move or rename any file in the vault with binary-safe file operations (file-only; not recursive directory moves). For safety, this tool requires confirmation of both source and destination paths.

Request:

{
  "name": "move_file",
  "arguments": {
    "oldPath": "Miro/attachments/Pasted image 20250812140124.png",
    "newPath": "assets/images/Pasted image 20250812140124.png",
    "confirmOldPath": "Miro/attachments/Pasted image 20250812140124.png",
    "confirmNewPath": "assets/images/Pasted image 20250812140124.png",
    "overwrite": false
  }
}

Response:

{
  "success": true,
  "oldPath": "Miro/attachments/Pasted image 20250812140124.png",
  "newPath": "assets/images/Pasted image 20250812140124.png",
  "message": "Successfully moved file from Miro/attachments/Pasted image 20250812140124.png to assets/images/Pasted image 20250812140124.png"
}

Safety Note: confirmOldPath must exactly match oldPath, and confirmNewPath must exactly match newPath, otherwise the move is rejected.

read_multiple_notes

Read multiple notes in a batch (maximum 10 files).

Request:

{
  "name": "read_multiple_notes",
  "arguments": {
    "paths": ["note1.md", "note2.md", "note3.md"],
    "includeContent": true,
    "includeFrontmatter": true,
    "prettyPrint": false
  }
}

Response (optimized, shortened field names):

{
  "ok": [
    {
      "path": "note1.md",
      "frontmatter": { "title": "Note 1" },
      "content": "# Note 1\n\nContent here..."
    }
  ],
  "err": [{ "path": "note2.md", "error": "File not found" }]
}

Field names:

• ok = successful reads
• err = failed reads

update_frontmatter

Update frontmatter of a note without changing content.

Request:

{
  "name": "update_frontmatter",
  "arguments": {
    "path": "research-note.md",
    "frontmatter": {
      "status": "completed",
      "updated": "2025-09-23"
    },
    "merge": true
  }
}

Response:

{
  "message": "Successfully updated frontmatter for: research-note.md"
}

get_notes_info

Get metadata for notes without reading full content.

Request:

{
  "name": "get_notes_info",
  "arguments": {
    "paths": ["note1.md", "note2.md"],
    "prettyPrint": false
  }
}

Response (optimized, returns array directly):

[
  {
    "path": "note1.md",
    "size": 1024,
    "modified": 1695456000000,
    "hasFrontmatter": true
  }
]

get_vault_stats

Get high-level vault statistics without reading note contents.

Request:

{
  "name": "get_vault_stats",
  "arguments": {
    "recentCount": 5,
    "prettyPrint": false
  }
}

Response (optimized):

{
  "notes": 1248,
  "folders": 76,
  "size": 18349210,
  "recent": [
    {
      "path": "Daily/2026-02-27.md",
      "modified": 1772188800000,
      "size": 2814
    }
  ]
}

Security Considerations

This MCP server implements several security measures to protect your Obsidian vault:

Path Security

• Path Traversal Protection: All file paths are validated to prevent access outside the vault
• Relative Path Enforcement: Paths are normalized and restricted to the vault directory
• Symbolic Link Safety: Resolved paths are checked against vault boundaries

File Filtering

• Automatic Exclusions: .obsidian, .git, node_modules, and system files are filtered
• Extension Whitelist: Only .md, .markdown, .txt, .base, and .canvas files are accessible by default
• Hidden File Protection: Dot files and system directories are automatically excluded

Content Validation

• YAML Frontmatter Validation: Frontmatter is parsed and validated before writing
• Function/Symbol Prevention: Dangerous JavaScript objects are blocked from frontmatter
• Data Type Checking: Only safe data types (strings, numbers, arrays, objects) allowed

Best Practices

• Least Privilege: Server only accesses the specified vault directory
• Read-Only by Default: Consider running with read-only permissions for sensitive vaults
• Backup Recommended: Always backup your vault before using write operations
• Network Isolation: Server uses stdio transport (no network exposure)

What's NOT Protected

• File Content: The server can read/write any allowed file content
• Vault Structure: Directory structure is visible to AI assistants
• File Metadata: Creation times, file sizes, etc. are accessible

⚠️ Important: Only grant vault access to trusted AI conversations. The server provides full read/write access to your notes within the security boundaries above.

Architecture

• server.ts - MCP server entry point
• src/frontmatter.ts - YAML frontmatter handling with gray-matter
• src/filesystem.ts - Safe file operations with path validation
• src/pathfilter.ts - Directory and file filtering
• src/search.ts - Note search functionality with content and frontmatter support
• src/uri.ts - Obsidian URI generation for deep links
• src/types.ts - TypeScript type definitions

Contributing

1. Fork the repository
2. Create a feature branch: git checkout -b feature-name
3. Make your changes and add tests
4. Ensure all tests pass: npm test
5. Submit a pull request

License

MIT