obsidian-emergent-mcp
Most obsidian MCPs I have seen just re impliment the basic features "read note, write note, search for note by tag/keyword", better, but basically just the same as standard claude code tools.
An MCP server that gives Claude Code deep access to your Obsidian vault. 28 tools for searching, navigating, and analyzing your knowledge graph — structural analysis, semantic retrieval, timeseries tracking, and more.
Your vault becomes a live knowledge base that Claude can read, search, and reason about.
Install
From inside your Obsidian vault, run:
curl -fsSL https://raw.githubusercontent.com/Gassandrid/obsidian-emergent-mcp/main/install.sh | bash
This installs two Python files into your vault's .claude/ directory, sets up .mcp.json, and installs dependencies. The MCP server starts automatically when you open the vault in Claude Code.
Requirements
- Python 3.10+
- An Obsidian vault (must have
.obsidian/directory) - Claude Code CLI
This tool is only really useful if you are Git versioning your vault, dont use this if you are not.
Smart connections will also add some more behavior. we dont use it for RAG but rather just cosine sim for related notes.
Tools
Search & Read
| Tool | What it does |
|---|---|
vault_rag | Primary retrieval. Free-text query → TF-IDF + embedding reranking → ranked results with excerpts |
vault_search | Substring search across titles, content, tags — for exact lookups |
vault_read | Read a note's content, tags, and link counts |
Graph Navigation
| Tool | What it does |
|---|---|
vault_backlinks | What links TO this note |
vault_forward_links | What this note links TO |
vault_context | Semantically related notes (embeddings + link proximity) |
vault_related | Structurally related notes (shared neighbors) |
vault_hubs | Most connected notes in the vault |
vault_orphans | Notes with no links in or out |
Timeseries (git-backed)
| Tool | What it does |
|---|---|
vault_attention_flow | What topics you're focused on — edits grouped by tag/folder with trend detection |
vault_topic_timeline | How a knowledge domain grows over time — edits bucketed by week/month/day |
vault_note_history | Full edit history of a specific note — commits, word counts, link counts at each snapshot |
vault_diff | Structural diff: new/deleted notes, link changes over N days |
vault_recent | Recently modified notes |
vault_knowledge_velocity | Which notes are being edited most |
vault_link_weights | Link strength via co-edit frequency |
Structure Analysis
| Tool | What it does |
|---|---|
vault_clusters | Community detection via label propagation |
vault_concept_gaps | Structural holes: missing notes, stub hubs, bridge gaps |
vault_broken_links | Broken wikilinks with inferred topic tags, filterable |
vault_maturity | Score notes by development maturity (word count, links, recency) |
vault_prereq_tree | Learning dependency tree from a root note |
vault_stats | Global metrics: note/link counts, tag distribution |
vault_tags | Tag frequency with example notes |
vault_suggest_links | Find semantically similar but unlinked note pairs |
vault_batch_tag | Preview bulk tag operations (dry-run only) |
vault_open_questions | Extract unresolved questions, TODOs, unchecked checkboxes |
Agent Memory
| Tool | What it does |
|---|---|
vault_journal_write | Record observations for future sessions |
vault_journal_read | Recall past observations, filterable by tag and time |
It is recommended to prompt Claude to update your vault CLAUDE.md file to better utilize these tools.