Suggest a server AI tools directory →

MCP servers for files and knowledge

Last verified: 2026-09

These servers put the model inside a folder, a notes vault, or a Drive. The useful question is not 'can it read files' — they all can — but which root you are willing to jail it in. Filesystem is a directory you pick. Memory is a JSON graph the model writes itself. Obsidian and Logseq are existing vaults. Google Drive is someone else's cloud.

What to look for

  • Start with one project folder, never $HOME. Secrets live next to code more often than people think.
  • Prefer a notes app server (Obsidian) if the knowledge already lives there — do not duplicate it into Memory.
  • Drive and local disk are different threat models. OAuth can see every file the account can; a folder jail cannot.

Side-by-side

ServerOfficial?Needs a secret?Best for
FilesystemYesNoRead and write files inside a folder you choose.
MemoryYesYesPersistent notes the model can store and recall across chats.
Google DriveYesYesList and read files from a Google Drive account.
ObsidianNoYesRead and search a local Obsidian vault.
LogseqNoYesRead a Logseq graph.
Apple NotesNoNoRead local Apple Notes on macOS.

All 6 servers

FAQ

Should I install Filesystem and Memory together?

Usually yes. Filesystem is the working tree; Memory is preferences you want to survive a new chat. Do not store tokens in Memory.

Is Obsidian safer than Filesystem?

Only if the vault is smaller than the folder you would have allowed. Same disk, tighter root.

Can the model see files outside the allowed directory?

Not through the official Filesystem server, unless you passed a parent of those files as a root.