Large Language Models

The Large Language Models (LLMs) Plugin gives Obsidian users access to LLMs through cloud providers (OpenAI, Anthropic, Google, and Mistral) and locally via GPT4All and Ollama. Models can be interacted with in the sidebar, main window, and a newly added, floating action button popup window.

Instructions

Installation: Download the plugin via the community plugin browser.

Using models from cloud-based providers:

1. In the plugin settings menu, enter an API key from one of the supported model providers
2. To interact with models, open one of the chat views using the newly added commands (see Commands section below)

Using models locally (GPT4All):

1. Download GPT4All
2. Download a model through GPT4All's model browser
3. In the setting menu of GPT4All, toggle on the "Enable Local Server" setting
4. Models downloaded via GPT4All will be selectable via the model switcher in each chat view

Using models locally (Ollama):

1. Install Ollama and pull the models you want to use
2. In the plugin settings, configure the Ollama host (default: http://localhost:11434)
3. Click "Discover Models" to detect your locally available models
4. Select an Ollama model from the model switcher in any chat view

Commands

Command	Description
Open modal	Opens the chat modal
Toggle FAB	Toggles the visibility of the Floating Action Button (FAB) used to open and close the chat popup window
Open chat in tab	Opens the chat window in a tab
Open chat in sidebar	Opens the chat window in the Sidebar

Models

Cloud-based:

Model provider	Status
OpenAI	Supported
Anthropic	Supported
Google	Supported
Mistral	Supported

Local:

Model provider	Status
GPT4All	Supported
Ollama	Supported

MCP Server

The plugin can run a built-in MCP (Model Context Protocol) server so Claude Desktop, or any other MCP-compatible client, can read and write this vault directly — no separate companion plugin required.

Setup:

1. In plugin settings, go to General → Features and enable MCP Server. A bearer token is generated automatically.
2. Open the new MCP Server tab to see the port, bearer token, and a ready-to-copy claude_desktop_config.json snippet.
3. Add that snippet to your Claude Desktop config file, then restart Claude Desktop:

{
  "mcpServers": {
    "obsidian-vault": {
      "url": "http://127.0.0.1:27125/mcp",
      "headers": {
        "Authorization": "Bearer <your-token>"
      }
    }
  }
}

Notes:

• The server binds to 127.0.0.1 only and requires the bearer token on every request — it is not reachable from other devices on your network.
• Available tools: list_files, read_file, search_vault (read-only, run immediately) and create_file, edit_file, move_file, delete_file (each pops a confirmation dialog in Obsidian before running, the same as in-app agent actions).
• Desktop only. Disabled by default — enabling it opens a local port only for as long as the toggle stays on.
• Regenerating the bearer token immediately invalidates the old one; update any connected MCP clients afterward.

Credits

• Johnny ✨
• Ryan Mahoney
• Evan Harris