obsidian -plugin

The Obsidian MCP Plugin connects your vault to AI assistants using the Model Context Protocol (MCP). Once the plugin is configured, your AI client can navigate your vault as a semantic graph, follow links between notes, perform semantic searches across all notes, and read, edit, or create content through the AI assistant. Core capabilities include graph traversal to explore relationships, concept discovery via semantic search, contextual awareness of where information lives, and intelligent synthesis that blends information from multiple notes. The plugin exposes eight semantic tool groups that let the AI perform file operations, content edits, viewing, graph navigation, workflow suggestions, dataview queries (if installed), bases/database views, and system information about the vault and server status.

To use it, connect your MCP-enabled AI client (such as Claude Desktop, Claude Code, or Continue.dev) to the local MCP server you run from Obsidian. The typical setup is to run the MCP server locally (e.g., via npx mcp-remote) and point the client to http://localhost:3001/mcp (or your configured port). Once connected, you can ask the AI to summarize notes, find connections between topics, fetch related ideas across the vault, or create and modify notes through the AI’s commands, all while respecting the vault structure and graph relationships.

Prerequisites:

• Obsidian installed and configured
• Access to an MCP-compatible AI client (e.g., Claude Desktop, Claude Code, or Continue.dev)
• Optional: BRAT plugin for beta testing integrations

Step-by-step installation:

1. Install Obsidian MCP Plugin

   • In Obsidian, go to Settings → Community plugins → Browse, search for "Semantic MCP" (or the Obsidian MCP Plugin name) and install.
   • Enable the plugin after installation.

2. Install BRAT (optional for beta testing)

   • If you want to use the BRAT-based beta plugin flow, install BRAT from https://github.com/TfTHacker/obsidian42-brat and add the beta plugin: aaronsb/obsidian-mcp-plugin.

3. Run the MCP server locally

   • Ensure you have Node.js available for npx usage.
   • Start the MCP server as shown in the configuration example (see mcp_config):

   npx mcp-remote http://localhost:3001/mcp

   • If you need TLS/HTTPS or authentication, configure the appropriate options (see examples in the README and adjust the args accordingly).

4. Configure your MCP client

   • In your AI client, point to the local MCP server URL (e.g., http://localhost:3001/mcp) and use the provided JSON configuration structure to register the server under mcpServers.

5. Verify the connection

   • Ensure the Obsidian vault is accessible and that the MCP client can establish a session with the local server. You should be able to prompt the AI to navigate notes, run semantic searches, or perform edits via tool commands.

Tips and common considerations:

• Port configuration: The default in examples uses 3001 for HTTP and 3443 for HTTPS. Ensure these ports are not blocked by a firewall.
• Authentication: If you enable API key protection, securely store and pass the API key via environment variables or client configuration. Example: {"env": {"AUTH": "Bearer YOUR_API_KEY"}} and include the header argument in the mcpServers config.
• TLS/SSL: If using HTTPS in development, you may need to disable TLS verification for local testing (e.g., NODE_TLS_REJECT_UNAUTHORIZED=0). Only disable in secure, trusted environments.
• Dataview and Bases: Dataview integration and Bases support are optional. Ensure related Obsidian plugins are installed to unlock those capabilities.
• Debugging: If the AI cannot access certain notes, check Obsidian permissions, vault paths, and ensure the MCP server has network access to the vault data.
• Updates: The MCP plugin and associated tooling may update; review release notes for breaking changes in the server interface or tool availability.