Category · Knowledge base MCP servers
Obsidian MCP Server: Give AI Read Access to Your Vault
The Obsidian MCP server turns a vault of markdown notes into something an AI assistant can actually search and reason over — tags, wikilinks, and frontmatter included, rather than treated as plain text. It is read-only by design, which is the right call for a knowledge base you have spent years building: the assistant can find and synthesise, but it cannot quietly rewrite your notes. This page covers what it exposes, why read-only matters, and the servers that complement it.
What this category covers
Knowledge-base MCP servers expose personal or team notes to an AI assistant as structured data. The Obsidian server reads a local vault and understands its conventions: frontmatter as metadata, wikilinks as a graph, tags as classification. That structural awareness is what separates it from pointing a generic filesystem server at the same folder — the assistant can follow a link chain or filter by tag rather than grepping text. If your goal is recall across sessions rather than search, start with the best memory MCP servers instead.
How it works
Point the server at your vault directory and register it in your MCP client. It provides full-text search across notes plus structured queries against tags, links, and frontmatter fields. Everything stays local — the vault is a folder of markdown files on your disk, and the server never uploads it. Note contents do enter the conversation when the assistant reads them, so the usual context-window discipline applies: ask for targeted searches rather than whole-vault sweeps. Teams keeping documentation online will want the Notion MCP server for the same job.
Buying guide
The read-only design is a feature, not a limitation — accept it before you work around it. If you do want write access, add a Filesystem MCP server scoped narrowly to the vault path rather than looking for a read-write Obsidian server, and keep write approval on so nothing changes without you seeing it; our file system MCP servers page covers how to scope that safely. For vaults over a few thousand notes, plain search starts to strain: pair Obsidian MCP with Chroma or another vector store so semantic queries work alongside keyword ones. And treat the vault as untrusted input in one specific sense — if you clip web content into notes, that content is now text the assistant will read, so a note containing instructions should be treated as data rather than a command.
The Tools, Ranked
Read-only access to tags, links, frontmatter, and full-text search across an Obsidian vault. Open source and local.
A knowledge graph that survives across conversations. Pairs well with Obsidian: the vault holds your written notes, Memory holds what the assistant has learned about your work.
Obsidian MCP is read-only, so if you want the assistant to create or edit notes, add a filesystem server scoped to the vault directory — and enable write approval.
The equivalent for teams that keep documentation in Notion rather than local markdown.
For very large vaults, embedding your notes into a vector store gives semantic recall that keyword search misses.
Managed persistent memory that improves with use, an alternative to Memory MCP if you want it hosted.
Crawls web sources into clean markdown, which is how most people get external reading into a vault in the first place.