Which integrations are available for this server?

Manages notes in an Obsidian vault, providing tools to read, write, and navigate markdown files with YAML frontmatter, including bidirectional link tracking (parents/children), semantic indexing, and hierarchical organization across five knowledge levels. Integrates with Ollama for local embedding generation, enabling semantic search, similarity matching, and automated metadata suggestions using locally-hosted language models. Integrates with OpenAI's API for cloud-based embedding generation, supporting semantic analysis, core sign calibration, and vector similarity operations for knowledge graph navigation.

NOUZ — MCP Server for Obsidian

Unified MCP server for semantic knowledge management in Obsidian. Works with local embedding models — no cloud required.

MIT License Python 3.10+ MCP

Quick Start

git clone https://github.com/KVANTRA-dev/NOUZ-MCP
cd NOUZ-MCP
pip install -r requirements.txt

export OBSIDIAN_ROOT=./vault
export EMBED_API_URL=http://127.0.0.1:1234/v1  # LM Studio or Ollama

python server.py

Connect via any MCP-compatible client (Claude Desktop, etc.) using stdio transport.

Modes

NOUZ has three operating modes. Set via MODE env var or config.yaml.

Mode	Description	Embeddings required
luca	Pure graph — YAML + links, no semantic classification	No
prizma	Graph + semantics — sign classification, core_mix, semantic bridges	Yes
sloi	Strict 5-level hierarchy — cycle detection, violation warnings	Yes

Start with LUCA if you just want graph navigation. Add PRIZMA/SLOI when you're ready to define your semantic cores.

Note Format

Each note uses YAML frontmatter:

---
type: module
level: 3
sign: T
status: active
tags:
  - research
parents:
  - Parent Note Name
parents_meta:
  - entity: Parent Note Name
    link_type: hierarchy
---

Your note content here.

Five Levels

Level	Type	Description
1	core	Top-level domain (e.g. "Mathematics")
2	pattern	Knowledge area within domain
3	module	Grouping within field
4	quant	Concrete atomic note
5	artifact	Leaf note, reference, log

The parents field is a list of note names (used by Obsidian/ExcaliBrain for wikilinks). The parents_meta field carries structured link metadata for NOUZ graph logic. Keep them in sync — NOUZ handles this automatically on write.

Semantic Etalons (Prizma / Sloi)

Etalons define the "axes" of your semantic space — reference texts that represent each knowledge domain. NOUZ classifies notes by cosine similarity to these vectors.

Copy config.template.yaml to config.yaml:

mode: prizma

etalons:
  - sign: T
    name: Technology
    text: "programming software architecture infrastructure machine learning neural networks algorithms"
  - sign: S
    name: Science
    text: "physics chemistry biology mathematics formal logic theorems cosmology quantum mechanics"
  - sign: H
    name: Humanities
    text: "philosophy psychology sociology history literature art culture ethics cognitive science"

Best practices:

Use 3–5 domains — more makes classification noisy
Each etalon should be a dense list of keywords, not a sentence
Run calibrate_cores after changing etalons, then check pairwise cosine similarity
Aim for similarity < 0.55 between any two etalons for clean separation

Tools

All modes

Tool	Description
`read_file(path)`	Read note with parsed YAML metadata
`write_file(path, content, metadata)`	Write note with YAML frontmatter
`list_files(level, sign, tags)`	Filter and list notes from vault
`index_all(with_embeddings)`	Index vault to SQLite DB
`embed(text)`	Get embedding vector from configured provider
`get_parents(path)`	Get parent nodes (files linking to this)
`get_children(path)`	Get child nodes (files this links to)
`format_entity_compact(path)`	Return entity formula: `(children)[entity]{parents}`

Prizma / Sloi only

Tool	Description
`calibrate_cores()`	Recompute etalon embeddings from config
`recalc_signs()`	Auto-assign signs to all indexed notes
`recalc_core_mix()`	Aggregate core_mix bottom-up through DAG
`suggest_metadata(path)`	Suggest sign, level, core_mix for a note
`suggest_parents(path)`	Suggest parent links by vector similarity

Environment Variables

Variable	Default	Description
`OBSIDIAN_ROOT`	`./obsidian`	Path to your Obsidian vault
`MODE`	`luca`	`luca` / `prizma` / `sloi` — overridden by config.yaml
`EMBED_PROVIDER`	`openai`	`openai` / `gigachat` / `ollama`
`EMBED_API_URL`	`http://127.0.0.1:1234/v1`	Embedding API endpoint
`EMBED_MODEL`	—	Model name (optional, uses provider default)
`EMBED_API_KEY`	—	API key (leave empty for local models)
`EMBED_ENABLED`	`true`	Set to `false` to disable embeddings entirely

Example Workflow

# 1. Start server
python server.py

# 2. Index your vault
index_all(with_embeddings=true)

# 3. Calibrate etalons (Prizma/Sloi)
calibrate_cores()

# 4. Classify all notes
recalc_signs()

# 5. Query
suggest_parents("path/to/note.md")
list_files(sign="T", level=4)

Links

Changelog

v2.1.1

Minor fixes and refactoring

v2.1.0

Initial public release

Install Server

A

security – no known vulnerabilities

A

license - permissive license

C

quality - C tier

How are these scores calculated?

Resources

GitHub Repository

Need Help?

Related Servers

Tools

View all tools

NOUZ MCP Server

NOUZ — MCP Server for Obsidian

Quick Start

Modes

Note Format

Five Levels

Semantic Etalons (Prizma / Sloi)

Tools

All modes

Prizma / Sloi only

Environment Variables

Example Workflow

Links

Changelog

v2.1.1

v2.1.0

Resources

Tools

Latest Blog Posts

MCP directory API