The problem
Every conversation with an AI assistant starts from zero. You re-explain your projects, repeat past decisions, and manually paste context from your notes. Your accumulated knowledge — the decisions you've made, the things you've learned, the connections between ideas — is invisible to your tools.
MindCache fixes this. It's an MCP server that connects your Obsidian vault directly to Claude Code, Claude Desktop, Cursor, and any MCP-compatible client. Your AI can search, read, and write to your vault as naturally as you do.
How it works
# Architecture
Obsidian Vault ←→ MindCache ←→ Claude Code
# MindCache runs locally. Nothing leaves your machine.
44 tools across 10 categories
Everything your AI needs to work with your knowledge.
Full-text search, natural language Q&A, find by tag, date, or related notes.
Read notes, sections, daily notes, frontmatter, and extract links.
Capture meetings, decisions, ideas, learnings, people, and references.
Append to daily notes, add tasks, generate reflection prompts.
Create notes from templates, append content, update sections, delete notes.
Add links, suggest connections, find broken links and orphan notes.
Add tags, update properties, rename and move notes.
Vault overview, tag distribution, folder structure.
Find checkbox tasks across the vault and mark them complete.
Bulk tag management, daily note history, and note export.
Get started
1. Run the setup. Point it at your vault folder.
$ npx @augmnt-sh/mindcache init
2. Add to Claude Code.
$ claude mcp add --scope user mindcache -- npx @augmnt-sh/mindcache@latest
That's it. No plugins needed. Claude Code can now search, read, and write to your vault.
Tool reference
All 44 tools, organized by what they do.
Find 7
search Full-text search with Obsidian operators ask Natural language Q&A across your vault find_related Notes connected via backlinks, tags, or content find_by_tag Filter notes by tags find_by_date Notes from a date range find_recent Recently modified notes find_mentions Where a term is mentioned Read 7
read_note Full note content read_section Content under a specific heading read_summary Note with metadata read_properties YAML frontmatter only read_today Today's daily note read_this_week Daily notes from current week extract_links All wikilinks and URLs from a note Remember 7
remember Quick capture to inbox remember_meeting Structured meeting notes remember_decision ADR-format decision log remember_idea Capture an idea remember_learning Save something learned remember_person Person or contact note remember_reference Bookmark with summary Journal 3
log Append to daily note log_task Add task to daily note reflect Reflection prompts Write 4
create_note Create note with optional template append_to_note Append to existing note update_section Replace section content delete_note Permanently delete a note Connect 4
add_link Insert wikilink between notes suggest_connections AI-powered link suggestions find_gaps Broken links (knowledge gaps) find_orphans Disconnected notes Organize 4
add_tag Add tags to a note update_properties Modify frontmatter rename_note Rename a note move_note Move to different folder Understand 3
vault_overview Stats, folders, tags list_tags All tags with counts get_vault_structure Folder tree Tasks 2
find_tasks Find checkbox tasks complete_task Mark task as done Vault 3
bulk_tag Add or remove tags across multiple notes daily_notes List all daily notes with stats export_note Export as clean markdown or JSON Who it's for
Anyone with an Obsidian vault and an AI assistant.
Developers
Document decisions, capture debugging sessions, onboard faster.
ask("Why did we choose JWT over sessions?") Researchers
Search across papers, connect ideas, track hypotheses.
find_related("Neural Plasticity.md") Writers
Organize research, track projects, find inspiration in past notes.
search("character arc draft") Students
Build knowledge bases across courses, review efficiently.
find_by_tag(["cs-301", "midterm"]) Professionals
Keep meeting notes, project context, and client knowledge accessible.
remember_meeting({title: "Q2 Review"}) Open source, local first
MindCache runs entirely on your machine. Your notes never leave your computer. No cloud, no accounts, no telemetry.
MIT licensed. Read the code, modify it, contribute to it.
$ npm install -g @augmnt-sh/mindcache
github.com/augmnt/mindcache →
Questions
What is MCP?
Model Context Protocol is an open standard for connecting AI assistants to external tools and data sources. Claude Code, Claude Desktop, and Cursor all support it.
Does my data leave my computer?
No. MindCache runs locally and connects directly to Obsidian on your machine. Nothing is sent to any external server.
Do I need Obsidian?
You need an Obsidian vault (a folder of markdown files). MindCache reads them directly from the filesystem — Obsidian doesn't even need to be running.
Which AI clients work with MindCache?
Any MCP-compatible client: Claude Code, Claude Desktop, Cursor, and others that support the protocol.
Is it free?
Yes. MindCache is free and open source under the MIT license.