MCP server connecting AI assistants to Obsidian vaults for reading, writing, searching, and managing notes
Eight tools covering content editing, global search, frontmatter management, tags, and folder operations
Requires Obsidian Local REST API plugin — works with Claude, Cursor, and other MCP clients
Pricing: Free forever
Best for: Individuals & startups
Pros & Cons
Pros
Comprehensive vault access — read, write, search, and manage metadata in one server
Atomic frontmatter operations prevent accidental data loss in YAML headers
Global search with date and path filters enables precise knowledge retrieval
Caching layer maintains functionality during temporary API disruptions
Works with any Obsidian vault through the Local REST API plugin
Cons
Requires the Obsidian Local REST API plugin to be installed and running
Obsidian desktop app must be open for the REST API to be accessible
No support for Obsidian-specific features like Canvas, graph view, or plugin commands
Large vaults may experience slow initial cache building
Key Features
Note reading in markdown or JSON format with creation/modification timestampsContent modification via append, prepend, or overwrite with auto-creation for new filesSearch-and-replace within notes supporting regex, case sensitivity, and whole-word matchingVault-wide global search with path filters, date filters, and paginated resultsAtomic YAML frontmatter management — get, set, or delete keys without rewriting filesTag management across both YAML frontmatter and inline note contentFolder operations including creation, renaming, moving, and deletion with hierarchy supportIn-memory caching with fallback search when the REST API is unavailable
Obsidian MCP is a Model Context Protocol server that connects AI agents and development tools to an Obsidian vault, enabling programmatic reading, writing, searching, and managing of notes through the Obsidian Local REST API plugin. It provides eight primary tools that cover the full spectrum of vault operations.
The server supports reading notes in markdown or JSON format with file statistics, modifying content through append, prepend, or overwrite operations, and performing search-and-replace with regex support. A global search tool conducts vault-wide queries with path and date filters and paginated results. Dedicated frontmatter and tag management tools handle YAML metadata atomically without rewriting entire files. The vault listing tool supports filtering by extension and name patterns.
An intelligent in-memory caching system improves performance and provides fallback search when the REST API is temporarily unavailable. The server bridges Obsidian into AI workflows, letting LLMs access and modify a knowledge base as part of research, writing, or coding tasks. Common use cases include automating note generation, batch-updating frontmatter across files, searching vaults through natural language, and maintaining linked knowledge graphs. It works with Claude Desktop, Claude Code, Cursor, and other MCP clients.
Does Obsidian need to be open for the MCP server to work?
Yes. The server connects through the Obsidian Local REST API plugin, which only runs when the Obsidian desktop app is open. If Obsidian is closed, the REST API endpoint is unavailable and the MCP server cannot access your vault.
Can the AI modify my existing notes?
Yes. The server supports append, prepend, and overwrite operations on any note, plus search-and-replace with regex support. It also provides atomic frontmatter editing that modifies YAML keys without rewriting the entire file, reducing the risk of accidental data loss.
Does it work with Obsidian Sync or third-party sync services?
The MCP server accesses the local vault on disk through the REST API. Changes made by the AI appear as normal file modifications, so Obsidian Sync, iCloud, Syncthing, or any file-based sync service will pick them up automatically on other devices.
Can it access Obsidian Canvas or graph view data?
No. The server works with markdown notes, frontmatter, tags, and folder structure only. Obsidian-specific features like Canvas, graph view, plugin commands, and the block reference system are not accessible through the Local REST API that the MCP server uses.