AI Second Brain
Obsidian MCP: Connect Claude to Your Vault Safely
Obsidian MCP explained: the servers that connect Claude to your vault, how to set each one up, when you don't need MCP at all, and how to keep notes safe.

Obsidian MCP connects Claude and other AI assistants to your Obsidian vault through the Model Context Protocol. A server exposes your notes as tools, such as search, read and append, and the assistant calls them mid-conversation.
It's the cleanest way to give Claude Desktop your notes. For Claude Code, you often don't need it at all.
I run Claude Code against my business vault every day, and I've tried the main MCP routes for the clients that can't touch files directly. Here's how each option works and how to set them up without exposing notes you'd rather keep private.
What is an Obsidian MCP server?
MCP is an open standard for connecting AI applications to external systems. Its maintainers describe it as a USB-C port for AI applications: one standard plug instead of a custom integration per tool.
An Obsidian MCP server is that plug for your vault. Once it's connected, an assistant can:
- Search notes by text or meaning
- Read a note, or part of one
- Create notes and append to daily notes
- Follow links and backlinks
- Update frontmatter properties
Because MCP is a shared standard, the same server works with Claude Desktop, Claude Code, Cursor, VS Code and other clients.
Obsidian MCP options compared
| MCP Connector plugin | mcp-obsidian | Claude Code IDE plugin | |
|---|---|---|---|
| How it runs | Inside Obsidian, on your machine | Python process via the Local REST API plugin | Inside Obsidian, local WebSocket |
| What Claude gets | Dozens of vault tools, including semantic search | Core file, search, patch and append tools | Your open file and current selection |
| Writes to the vault | Yes, per token permissions | Yes | No, read-only context |
| Extra setup | None beyond the plugin | Python, uv and an API key | None, connect with /ide |
| Best for | Claude Desktop and multi-client setups | People who prefer a standalone server | Claude Code users who want editor context |
Option 1: the MCP Connector plugin
The MCP Connector community plugin is the most complete option I've found. Its server runs inside Obsidian and binds to your own machine only, so nothing goes through a third-party cloud. Semantic search runs on-device with a local embedding model, and it doesn't need an API key.
Its best security feature is per-client tokens. Each connected app gets its own token with its own tool list, so a chat client can see a small read-only set while Claude Code gets the full toolbox.
Setup takes a few minutes:
- In Obsidian, go to Settings, Community plugins, Browse, and install MCP Connector.
- Open its Access control settings. A new vault already has a default token.
- For Claude Desktop, use the bundle file it offers. For other clients, copy the generated configuration.
- Ask the assistant to call the server info tool to confirm the connection.
It requires Obsidian 1.7.2 or newer.
Option 2: mcp-obsidian
mcp-obsidian is a popular open-source server written in Python. It talks to Obsidian through the Local REST API community plugin, so you install that plugin first and copy its API key.
It offers the essentials: list files, get file contents, search, patch content under a heading or block, append content and delete files. You configure it in your client with an environment variable for the API key, plus host and port if they differ from the defaults.
Choose it if you want the server outside Obsidian, or you already use the Local REST API for other automations.
Option 3: Claude Code without MCP
If you use Claude Code, start here before installing anything. Open a terminal in your vault and run claude. Claude Code reads and writes the files directly, which is more capable than most MCP tools for bulk work like restructuring folders or fixing links across hundreds of notes.
Two optional add-ons fill the gaps:
- The Claude Code IDE plugin shares the note you have open and your selection with Claude Code through a local, read-only bridge. After installing it, run
/idein Claude Code and pick Obsidian. Then "rewrite the second paragraph" just works. - Obsidian CLI lets Claude Code run Obsidian commands, such as renaming a note with automatic link updates.
You can still add an MCP server to Claude Code for semantic search. Remote servers use claude mcp add --transport http, and the Claude Code MCP docs cover local servers and scopes.
How to use Obsidian MCP safely
Connecting AI to your notes is connecting AI to your thinking. A few rules I follow:
- Know what leaves your machine. A local server keeps the connection local, but everything Claude reads is sent to Anthropic to generate the answer. The Claude Code IDE plugin's own FAQ says exactly this.
- Split vaults by sensitivity. Keep journals, health notes and credentials in a vault no AI client can reach.
- Give the fewest tools possible. Chat clients rarely need delete or bulk replace. Use per-token tool lists when your server supports them.
- Be wary of fetch tools. Anthropic's MCP documentation warns that servers fetching external content can expose you to prompt injection.
- Keep the vault in Git. If an assistant makes a bad bulk edit, you can roll it back.
Which Obsidian MCP setup should you choose?
- Claude Desktop user: MCP Connector, with a read-mostly token.
- Claude Code user: run it in the vault, add the IDE plugin, and add MCP only if you want semantic search.
- Several AI tools on one vault: MCP Connector with one token per client.
- Building an AI-maintained knowledge base: Claude Code with a schema file and skills. See how to build an Obsidian second brain with Claude Code.
For a broader look at the MCP servers worth connecting to Claude Code beyond Obsidian, read the best MCP servers for Claude Code. And if you'd like the whole system designed around your work, that's what I build in the AI Brains.
Frequently asked questions
What is Obsidian MCP?
It's any Model Context Protocol server that connects an AI client to an Obsidian vault. The server offers tools such as searching notes, reading files and appending content, which assistants like Claude can call during a conversation.
How do I connect Claude to Obsidian?
For Claude Code, the simplest way is to run claude inside your vault folder. For Claude Desktop or other clients, install an MCP server such as the MCP Connector community plugin and add its configuration to the client.
Is Obsidian MCP safe?
It's as safe as the permissions you grant. Prefer servers that run locally, give each client only the tools it needs, keep destructive tools off for chat clients, and remember that whatever Claude reads is sent to the model provider.
Do I need an MCP server to use Claude Code with Obsidian?
No. Claude Code reads and writes files directly. An MCP server mainly adds conveniences like semantic search, Obsidian-aware link handling and access for clients that can't touch your filesystem.
Sources
Christian Monge
Builds AI second brains with Obsidian and Claude Code, co-founded the competitive-intelligence SaaS Pengu Insights, and trains teams in English and Spanish from Madrid.
More about meAI Brains
Want this running on your own material?
I build Obsidian + Claude Code brains around your documents, calls and research, with an evidence ledger so every number has a source.
See the AI BrainsKeep reading

AI Second Brain
AI Note Taking App: How to Choose One You Keep
How to pick an AI note taking app: the four jobs they do, why file ownership decides the long game, and the setup that makes notes usable by AI later.

AI Second Brain
Karpathy's LLM Wiki: How to Build One in Obsidian
Karpathy's LLM wiki explained: why a persistent markdown wiki beats RAG, the three layers and three operations, and a step-by-step build with Obsidian.

AI Second Brain
Obsidian AI: The Setup That Turns Notes Into Answers
Obsidian AI explained: Obsidian has no built-in AI, so here are the plugins and agent setups worth using, how they differ, and which one fits your notes.