Obsidian MCP Server
Let AI agents read, write, search, and manage notes in your Obsidian vault
The Obsidian MCP server is a community implementation that gives AI agents direct access to an Obsidian vault, letting them read notes, create and update files, search by content or tag, and navigate the folder structure. Because Obsidian vaults are plain Markdown files on your local file system, the MCP server works by reading and writing those files directly, making setup simple and the integration reliable.
Obsidian vaults are where a lot of people keep their actual thinking: notes from meetings, project research, writing drafts, reference material, personal knowledge that does not fit anywhere else. The problem is that an AI agent cannot use any of that unless it can read it.
The Obsidian MCP server fixes this. It gives an agent direct access to your vault, so instead of copying and pasting notes into a conversation you can say "look up what I wrote about X" or "add a note about this to my vault" and have it happen. For people who use Obsidian as an active working environment, this is the integration that makes AI agents genuinely useful rather than a separate context they have to translate to.
How the server works
Obsidian vaults are plain Markdown files in a folder on your file system. That is the entire technical foundation, and it is why the MCP server is both simple and reliable: it reads and writes those files directly. There is no API to call, no plugin to install in Obsidian, no sync service to configure.
This architecture has a practical benefit: the server works whether or not Obsidian is currently open. Changes the agent makes (new notes, updates to existing ones) appear in Obsidian the next time you open or refresh the vault. Changes you make in Obsidian are immediately visible to the agent on the next read. There is no sync step and no separate database to keep in sync with your files.
The tradeoff is that the server operates at the file level, not through Obsidian's internal model. It does not use Obsidian's graph, it does not trigger Obsidian plugins, and it does not understand custom plugin-added functionality beyond frontmatter metadata. For most use cases this is fine. For workflows that depend heavily on specific plugins, test carefully.
What the server exposes
read_note. Fetches the content of a specific note by path, returning the Markdown content including frontmatter. The agent gets the actual note text, not a summary or excerpt.
create_note and update_note. Write operations for creating a new note or updating an existing one. The agent can write the content in Markdown format, including frontmatter fields. For journaling, meeting notes, or any workflow where the agent helps generate content for the vault, these are the core tools.
search_vault. Full-text search across all notes in the vault. Accepts a query string and returns matching notes with the matching content sections. For vaults with hundreds or thousands of notes, this is the primary discovery mechanism.
list_notes. Lists notes in a specific folder or across the entire vault. Can be filtered by folder path to scope the listing to a section of the vault.
find_by_tag. Returns notes that have a specific tag in their frontmatter or inline tags in the body. This maps directly to how Obsidian users organize notes with tags, making it easy for the agent to find thematically related content.
read_frontmatter. Extracts just the frontmatter metadata from a note without loading the full body. Useful for scanning note metadata at scale without the overhead of loading entire files.
Setup
The setup is simpler than most MCP servers because there is no API key or cloud service involved. You need the path to your Obsidian vault folder.
Your vault path is wherever you created the vault. On macOS it is typically somewhere like /Users/yourname/Documents/ObsidianVault. Open Obsidian and hover over the vault name in the bottom left to see the full path.
Add to Claude Desktop at ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"obsidian": {
"command": "npx",
"args": [
"-y",
"@smithery/mcp-obsidian",
"--vault",
"/Users/yourname/Documents/ObsidianVault"
]
}
}
}Replace the vault path with your actual vault location. Use an absolute path.
For Claude Code, the config lives at ~/.claude/mcp.json with the same structure.
If you have multiple vaults, you can add the server block multiple times with different names and different --vault paths:
{
"mcpServers": {
"obsidian-work": {
"command": "npx",
"args": ["-y", "@smithery/mcp-obsidian", "--vault", "/Users/yourname/Work/Notes"]
},
"obsidian-personal": {
"command": "npx",
"args": ["-y", "@smithery/mcp-obsidian", "--vault", "/Users/yourname/Personal/Notes"]
}
}
}Real use cases
Retrieving your own research. When you are working on something and have notes about it, the agent can search the vault and pull the relevant content. "What do I have on topic X?" becomes an actual query with actual results rather than a manual search through your note hierarchy.
Meeting and session notes. After a meeting, describe what happened and have the agent create a structured note in your vault: date, attendees, discussion points, action items. The note lands in the right folder with consistent frontmatter. Over time, search and tag tools make these notes findable by project or person.
Writing drafts. Use the agent to draft longer pieces (articles, proposals, reports) and write them directly into the vault. The agent can create the note, add appropriate frontmatter, and link to related notes if you have an established naming convention. Drafts stay in the vault rather than disappearing into conversation history.
Knowledge synthesis. If your vault has notes on a topic from multiple sources, the agent can search for them all, read the content, and synthesize a consolidated summary. "What have I written about X over the last year?" is the kind of question that would normally require manual review of multiple notes.
Daily notes and journaling. For people using Obsidian's daily notes feature, the agent can read or update today's note, add items to it, or search recent daily notes for context on an ongoing task. This is the workflow where having the agent directly in the vault is most different from copying and pasting.
Frontmatter management. If you want to retroactively add tags, update the status field, or add publication dates to a set of notes, the agent can iterate through the matching files and update the frontmatter. A task that would take an hour manually takes a few minutes as a conversational request.
Cross-note linking suggestions. The agent can read a new note you are writing, search the vault for related content, and suggest which existing notes might be worth linking. Obsidian's graph becomes more useful when it is actually dense with links.
Vault security and privacy
Your Obsidian vault likely contains private writing. Before setting up the MCP server, think about what you are comfortable having an AI agent read:
The agent reads everything in the vault path you configure. There is no per-folder access control in the current implementation. If you point it at your full vault, the agent can see all of it. If you have sensitive material (medical notes, financial information, private correspondence) in a subfolder, consider creating a separate vault for content you are comfortable sharing with the agent, or accept that the agent will have access to everything.
Content leaves your machine. When the agent reads a note to answer a question, the content of that note is sent to the model provider (Anthropic, or whoever is running the model). The privacy implications of this are the same as pasting your notes into a chat window. For most vault content this is fine; for genuinely sensitive material, be deliberate about which vault you configure.
The agent can write to your vault. Create and update operations modify files on disk. There is no confirmation step. Use specific prompts for write operations, and keep a backup strategy for your vault (most people already do this through iCloud, Dropbox, or a git repository).
Community alternatives
Because there is no official Obsidian MCP server, several community implementations exist. The smithery-ai implementation covered here is widely used and actively maintained. Other notable options:
mcp-obsidian by calclavia. Another community implementation with a slightly different tool surface. Worth comparing if the smithery version does not cover your workflow.
Obsidian Local REST API plugin. Not an MCP server itself, but a community Obsidian plugin that exposes the vault through a local REST API. Several MCP servers are built on top of this plugin, which gives them access to Obsidian's internal model (including graph data and plugin state) rather than just the raw file system. If you need deeper integration with Obsidian's features, implementations that use the REST API plugin are worth exploring.
All of these are community projects. Check the GitHub repositories for maintenance status and recent activity before building a workflow around any of them.
Pairing with other servers
Obsidian MCP is the memory layer for your personal knowledge. It pairs well with any workflow that benefits from that context.
Combined with Tavily MCP or Brave Search MCP, the agent can search the web for new information and write it into the vault in a structured format. Research sessions create vault entries automatically.
Combined with Filesystem MCP, the agent can handle files outside the vault (PDFs, spreadsheets, exported data) alongside vault notes. The two together give coverage of your entire file-based knowledge, not just the Markdown notes.
Combined with GitHub MCP, a development agent can read technical notes from the vault while working on code, or write implementation notes and decisions back to the vault as part of a coding session.
Combined with Memory MCP, you get two distinct memory layers: the Memory server's conversation-level graph for short-term agent context, and the Obsidian vault for long-term personal knowledge. They serve different purposes and work together.
Bottom line
The Obsidian MCP server is the bridge between your personal knowledge base and the AI agents you work with. For people who use Obsidian actively, the gap it closes is real: instead of translating between your notes and your conversations, the agent can read and write directly.
The community implementation is solid for core vault operations. The setup is genuinely simple (just a vault path, no API key). The main consideration is privacy: be deliberate about which vault you expose and know that the agent can see everything in it.
For knowledge workers whose vault contains the context that would make AI assistance actually useful, this is one of the higher-value MCP integrations to configure.
Features
- Read any note from the vault by path
- Create and update Markdown notes
- Search vault content by keyword or regex
- List notes by folder or tag
- Read and write frontmatter metadata
- Navigate folder structure and file hierarchy
- Find notes by tag combinations
- Works with any Obsidian vault without a plugin requirement
How to set up the Obsidian MCP Server MCP server
- Locate your Obsidian vault path (the folder containing your .md files)
- Add the server block to your Claude Desktop or Claude Code MCP config
- Set the vault path in the server arguments
- Restart your MCP client and verify vault tools appear in the tool list
- Test by asking the agent to list notes or search for a keyword