Obsidian MCP Server — Claude Code MCP server: features and setup
Connects MCP clients to an Obsidian vault that agents can use as shared memory. It searches notes by meaning and keywords and works with wikilink connections.
MIT ★ 20 updated 2026-10-29
Open on GitHubWhat Obsidian MCP Server does
This server gives MCP clients access to Markdown files in an Obsidian vault. An agent can search by keyword or meaning, follow wikilinks, and use graph operations across notes. The memory stays in readable files that you can open and edit in Obsidian.
Before applying an edit, the server can check expected_hash and reject a change based on an outdated read. Publishing is atomic, deleted files go to .trash/, and dry_run returns a diff before changes are applied. The listed stack is Python 3.12, FastAPI, and PostgreSQL with pgvector. It is self-hosted, so you need to deploy the server and configure a client connection; the README shows an HTTP connection with an authorization token for Claude Code.
What it helps with
- search vault notes by meaning or keywords
- follow wikilink connections between notes
- keep agent memory in ordinary Markdown files
- reject edits based on stale note contents
Example prompts for the agent
- Search my vault for notes about this project and follow their wikilinks.
- Save a summary of this session in the right Obsidian note.
- Show me the diff for this note change before applying it.
How to install Obsidian MCP Server
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Obsidian MCP Server
How do I connect Obsidian MCP Server to Claude Code?
The README demonstrates an HTTP connection that uses the server address and an authorization header. Deploy the server, then configure the client with the connection details shown in the README.
What does Obsidian MCP Server do?
It gives MCP clients access to Obsidian notes for searching, reading, and writing. Search supports keywords and semantic queries, and the server can work with the wikilink graph.
How does it prevent overwriting a changed note?
The expected_hash check rejects an edit if the note has changed since it was read. You can inspect a diff with dry_run first; deleted files are moved to .trash/.