LabscoMCP early access ↗
tomjrworksMCP SERVER

Taproot for Obsidian

Connect an AI client to your Obsidian vault — save anything into it from anywhere, and get cited answers back out of it.

by tomjrworks ↗Note-Taking & Personal Knowledge BasesCommunity

LABSCO SUMMARY

Your vault as the assistant's working memory.

Notes go in from anywhere and come back as cited answers, and every answer can be written back, so the vault compounds instead of staying a pile of clippings. It reads the folder structure and wikilink conventions you already use rather than imposing its own.

What it is

An MCP server that gives Claude, ChatGPT, Cursor or Windsurf read and write access to an Obsidian vault, sandboxed to that directory. It ships on npm as synapse-obsidian and presents itself as Taproot: nine taproot_ tools for the knowledge-base workflow and six garden_ tools for direct file access. The vault is just a folder of markdown, so Obsidian does not need to be running.

What you get

  • A guided first run that scans the vault and offers three routes — use the structure you already have, scaffold a knowledge base, or describe your own — then saves the answer (taproot_plant, taproot_till, taproot_sow)
  • Anything saved into the vault from anywhere: hand it a URL and the page is fetched, converted to markdown and filed (taproot_seed)
  • Raw sources turned into summaries, concept pages and entity pages, cross-linked with wikilinks — one at a time or every unprocessed source at once (taproot_water, taproot_cultivate)
  • A question researched across the whole knowledge base, answered with citations (taproot_harvest)
  • A full vault overview with suggested next actions, and a health check for broken links, orphan pages, missing frontmatter and stale content (taproot_status, taproot_prune)
  • Direct file access underneath all of it: read a file, create or update one, list a directory, full-text search, count files, read a file's YAML metadata (garden_read, garden_plant, garden_survey, garden_forage, garden_measure, garden_tag)
  • A --http mode so a browser-based client can reach a vault on your own machine through a tunnel

Requirements

Node.js 18 or newer, and the path to your vault — or any folder of markdown — passed as an argument to npx -y synapse-obsidian. For Claude.ai or ChatGPT, run it with --http --port 3777 and expose it through a tunnel, then add the tunnel's /mcp URL as a custom integration. Access is sandboxed to the vault directory and nothing outside it.

Setup effort

One command — npx -y synapse-obsidian <YOUR_PATH_TO_OBSIDIAN_VAULT>