Working with Obsidian and Logseq: PKM Tools and Claude

Who this is for: Researchers who use Obsidian or Logseq as a personal knowledge management (PKM) system and want to integrate Claude into their note-making practice. This document covers both tools — Obsidian via direct file access, Logseq via a dedicated MCP server with richer graph-level integration.

The key insight: your vault is already Claude-readable

Both Obsidian and Logseq store notes as plain markdown files in a folder on your disk. This is the same format Claude Code works with natively. There is nothing special to configure for basic use:

cd /path/to/your/vault
claude

You are now inside your vault. Claude can read any note by name, search across all notes, write new notes, and modify existing ones — because they are just files.

> Read my note on "Cathar perfecti" and find all notes that link to it.

> Search all notes in this vault for mentions of "Fournier Register".
  List which notes mention it and what context they use it in.

No plugins. No export. No upload. The vault is the project folder.

Obsidian + Claude Code

What Obsidian is

Obsidian is a markdown-based note-taking application. Notes are stored as .md files in a folder (the vault). Obsidian adds features on top: wiki-links ([[note name]]), backlinks, graph view, tags, properties (YAML frontmatter), and a plugin ecosystem.

From Claude's perspective: a vault is a folder of markdown files with some conventions — wiki-link syntax, frontmatter headers, tag notation (#tag). Claude understands all of these without special configuration.

Setting up

No MCP server needed. Just navigate to your vault and launch Claude:

cd ~/Documents/ObsidianVault
claude

If you want Claude to know about your vault structure, add a CLAUDE.md:

# My Obsidian vault

Structure:

- /notes/         — permanent notes (Zettelkasten)

- /literature/    — reading notes, one per source

- /daily/         — daily notes (YYYY-MM-DD.md format)

- /projects/      — project-specific notes

- /inbox/         — rough notes to process

Conventions:

- Wiki-links use [[exact note title]]

- Tags: #to-process #permanent #literature #question

- Frontmatter: title, date, tags, source (for literature notes)

- Sources are linked to Zotero keys where available

Practical workflows

Backlinking: find and add missing links

> Read my note [[Social networks in Catharism]].
  Find all other notes in the vault that mention the same persons or concepts
  but don't yet have a wiki-link to this note.
  List them with the relevant passage. I will decide which links to add.

Processing inbox notes into permanent notes

> Read all notes in /inbox/ tagged #to-process.
  For each one: suggest whether it should become a permanent note,
  a literature note, or a project note — based on its content.
  For notes that should become permanent: suggest a title and
  which existing notes it could link to.
  Do not move or modify anything; just give me the assessment.

Finding orphaned notes

> List all notes in /notes/ that have no incoming wiki-links
  (no other note links to them).
  For each orphan: suggest 2–3 existing notes it could connect to,
  based on content similarity.

Vault-wide synthesis

> Read all notes tagged #question in this vault.
  What recurring questions appear across my notes?
  Are there questions I keep returning to without resolving?
  Group them by theme.

Frontmatter consistency audit

> Check all notes in /literature/ for frontmatter consistency.
  Each note should have: title, author, year, source, tags.
  List any notes missing required fields.
  Do not modify files — just report.

Building a Map of Content

> I want to create a Map of Content for the topic "medieval inquisition procedure".
  Search the vault for all notes related to this topic.
  Draft a new note called [[MOC - Medieval Inquisition Procedure]] that:

  - Lists the relevant notes grouped by subtopic

  - Uses wiki-links throughout

  - Has a brief introductory paragraph describing the intellectual territory
  Save it to /notes/MOC - Medieval Inquisition Procedure.md

Daily note processing

> Read today's daily note (/daily/2025-03-12.md).
  Extract any tasks, ideas, or references that should go elsewhere.
  For tasks: list them. For ideas that should become notes: suggest titles.
  For references to existing notes: check if the relevant notes exist.

What to protect: your words stay yours

The critical rule from the Zettelkasten tradition applies here: Claude organises and connects; it does not write your notes.

If Claude generates the content of your permanent notes, you lose the intellectual work of formulation. The act of writing a permanent note — finding your own words for an idea — is part of how you understand it. Claude is useful for:

It is not useful for writing permanent notes on your behalf, even if asked.

Logseq + Claude Code: MCP integration

What Logseq is

Logseq is an outliner-based PKM tool, also markdown-based, but with a different data model: notes are composed of blocks (bullet points), and the fundamental unit is the block, not the file. Logseq has strong journal integration — each day has a journal page, and most notes grow from journal entries.

Unlike Obsidian, Logseq has a dedicated MCP server that gives Claude graph-level awareness — not just file reading, but understanding of Logseq's structure: pages, blocks, backlinks, journal entries, queries.

Setting up the Logseq MCP

Install the Logseq MCP server:

There are two components: a search server and a write server. Add both to ~/.claude/settings.json:

{
  "mcpServers": {
    "logseq-search": {
      "command": "npx",
      "args": ["-y", "logseq-mcp-search"],
      "env": {
        "LOGSEQ_GRAPH_PATH": "/path/to/your/logseq/graph"
      }
    },
    "logseq-write": {
      "command": "npx",
      "args": ["-y", "logseq-mcp-write"],
      "env": {
        "LOGSEQ_API_TOKEN": "your-logseq-api-token",
        "LOGSEQ_HOST": "http://localhost:12315"
      }
    }
  }
}

The write server requires Logseq to be running and the API server enabled in Logseq settings (Settings → Advanced → Enable HTTP APIs server).

For read-only use (search and analysis without writing back), the search server alone is enough and does not require Logseq to be running.

Allow tools in permissions:

{
  "permissions": {
    "allow": [
      "mcp__logseq-search__*",
      "mcp__logseq-write__*"
    ]
  }
}

What the MCP gives Claude

The Logseq MCP exposes graph-level operations beyond simple file reading:

Capability What it means
searchPages / searchDocuments Semantic and keyword search across the graph
getPage / getBlock Retrieve any page or block by name/ID
getBacklinks Find all pages that reference a given page
getAllPages List all pages in the graph
getJournalSummary Summarise journal entries by date range
analyzeJournalPatterns Find patterns across journal entries
suggestConnections Suggest which pages should be linked
findKnowledgeGaps Identify underexplored areas in the graph
analyzeGraph High-level graph structure analysis
addJournalEntry / addJournalBlock Write to today's journal
addNoteContent Add content to any page
createPage Create a new page

This is substantially richer than file access — Claude understands that your graph has structure, not just files.

Practical Logseq workflows

Journal summary and pattern analysis

> Summarise my journal entries from the last two weeks.
  What topics have I been thinking about most?
  What tasks keep appearing without being resolved?
  What questions have I noted that I haven't followed up on?

Claude reads your journal entries and produces a structured summary — useful for weekly reviews without reading two weeks of daily notes yourself.

Finding knowledge gaps

> Analyse my Logseq graph.
  What topics do I have many notes on but few connections between?
  What areas seem underexplored relative to how often I reference them?
  What should I read or think about more?

Suggesting connections

> Look at my page [[Kinship networks in Languedoc]].
  Which other pages in my graph should be connected to it
  but currently aren't linked?
  Explain why each connection would be valuable.

Daily capture and processing

> Add to today's journal:
  Meeting with [name] — discussed progress on Montaillou chapter.
  Key decision: focus the network analysis on the 1308-1320 period.
  TODO: re-run centrality measures with updated data.

Claude writes the entry to your Logseq journal directly. Useful for capturing notes from a conversation while keeping your hands free, or for logging research decisions in a session.

Processing a backlog of unlinked references

> Get all pages in my graph that have the tag [[unprocessed]].
  For each one: read the content and suggest which existing pages
  it should be linked to. Show me the suggestions grouped by page.

Cross-graph search for a research question

> Search my Logseq graph for everything related to
  "economic relations between perfecti and credentes".
  Include both pages with that topic and journal entries that mention it.
  Synthesise what I have written about this across all these notes.

Graph structure analysis

> Analyse the overall structure of my Logseq graph.
  Which pages are the most connected hubs?
  Which pages are isolated?
  What does the graph structure suggest about where my thinking is developed
  vs. where it is sparse?

Obsidian vs. Logseq for Claude integration

Obsidian Logseq
Claude access method File system (always works) File system + MCP (richer)
Graph-level queries Manual (Claude reads files) Native via MCP
Journal integration File per day; Claude reads them MCP understands journal structure
Write back to notes Claude writes .md files MCP writes blocks/entries natively
Setup complexity None — just cd into vault MCP install + Logseq API setup
Works without app running Yes (files are always there) Partially (search yes; write needs app)
Best for Established Obsidian users Logseq users wanting deep integration

If you use Obsidian, the file-access approach is simple and powerful. The MCP plugin ecosystem for Obsidian is developing but not as mature as the Logseq MCP.

If you use Logseq, setting up the MCP is worth the one-time effort — the graph-level awareness enables workflows that file reading alone cannot.

The Zotero + PKM + Claude triangle

The full integrated workflow for a research project:

Zotero          ←→   Claude    ←→   Obsidian / Logseq
(references)         (reasoning)      (notes and thinking)

- Search library     - Find connections   - Store atomic notes

- Retrieve metadata  - Synthesise across  - Link concepts

- Add notes          - Process inbox      - Journal research

- Cite sources       - Fill gaps          - Map knowledge

Claude operates at the centre — reaching into Zotero for sources, reaching into your vault for notes, and helping you connect the two. Neither Zotero nor your vault changes fundamentally; Claude becomes the active interface between them.

Practical invocation:

> Search my Zotero library for sources on [topic].
  Then search my Logseq graph for notes where I have engaged with this topic.
  Tell me: what sources in Zotero am I not yet engaging with in my notes?
  What ideas in my notes are not yet grounded in specific sources?

Related