knowledge-graph

name	knowledge-graph
description	Query the Obsidian vault as a live knowledge graph — PageRank, community detection, semantic search, path-finding, and vault writes via MCP.
tags	["skill","mcp","knowledge","obsidian","graph","search"]
triggers	["knowledge graph","use knowledge graph","run knowledge graph"]

Knowledge Graph — Vault as Queryable Graph

MCP server: knowledge-graph Source: C:\Users\User\tools\knowledge-graph Vault: C:\Users\User\Business-Empire-Agent (2,117 nodes / 3,725 edges / 696 communities at last index) Data dir: C:\Users\User\tools\knowledge-graph\data

What This Gives You

The Obsidian vault is parsed into a SQLite-backed graph with:

• PageRank centrality — which notes are most structurally important
• Betweenness centrality (bridge nodes) — which notes connect otherwise separate clusters
• Louvain community detection — automatically grouped topic clusters (696 detected)
• Semantic search — transformer embeddings (lazy-loaded on first use)
• Full-text search — BM25-style keyword search, instant
• Path-finding — shortest connecting path between any two nodes
• Vault writes — create new nodes, annotate existing ones, add wiki-links

Tools Reference

Read Tools

Tool	When to Use	Key Params
kg_search	Find relevant notes by meaning or keyword	query, fulltext (bool), limit
kg_node	Inspect a single note — metadata, links, content	name (fuzzy), brief (bool), maxContentLength
kg_neighbors	Explore what connects to a note	name, depth (default 1)
kg_subgraph	Get a local neighborhood as a self-contained graph	name, depth
kg_paths	Find connecting route between two notes	from, to, maxDepth
kg_common	Find shared connections between two notes	nodeA, nodeB
kg_central	Top notes by PageRank (most linked-to)	limit, community (optional filter)
kg_bridges	Top bridge nodes by betweenness centrality	limit
kg_communities	List all Louvain community clusters	—
kg_community	Get members of a specific community	id (number or label)

Write Tools

Tool	When to Use	Key Params
kg_index	Re-index vault after large changes	resolution (Louvain, default 1.0)
kg_create_node	Create a new markdown file in the vault	title, directory, content, frontmatter
kg_annotate_node	Append content to an existing note	name, content
kg_add_link	Add a wiki-link from one note to another	source, target, context

When to Use Knowledge Graph vs Other Tools

Task	Use
"What notes relate to PropFlow revenue?"	kg_search (semantic)
"Find all notes tagged [skill]"	kg_search fulltext with tag query, or Grep
"What does STATE.md link to?"	kg_node "STATE"
"How does SOUL.md connect to ACTIVE_TASKS?"	kg_paths
"Which notes are most central to the vault?"	kg_central
"Which notes bridge disconnected clusters?"	kg_bridges
"What topic cluster does client-health belong to?"	kg_node → check community in frontmatter result
Find a specific string in a file	Grep (exact match is faster than graph)
Read a specific file's content	Read tool (direct, no overhead)
Search mem0 knowledge graph	Memory MCP search_nodes (different graph — personal facts, not vault)

Decision: Knowledge Graph vs Memory MCP vs Grep

Query is about vault STRUCTURE or RELATIONSHIPS?
  YES → knowledge-graph MCP

Query is about CC's personal facts, preferences, or past decisions?
  YES → Memory MCP (search_nodes)

Query is a known string/pattern in a known file?
  YES → Grep (faster, exact)

Query is "find notes conceptually similar to X"?
  YES → kg_search (semantic, no fulltext flag)

Query is "find notes containing the word X"?
  YES → kg_search (fulltext: true) OR Grep

Example Queries

# What notes are semantically related to MRR growth?
kg_search { query: "MRR growth revenue pipeline", limit: 10 }

# Show me STATE.md with its connections
kg_node { name: "STATE", brief: true }

# How does the SOUL connect to ACTIVE_TASKS?
kg_paths { from: "SOUL", to: "ACTIVE_TASKS", maxDepth: 4 }

# What are the most central notes in the vault?
kg_central { limit: 10 }

# Which notes bridge disconnected clusters?
kg_bridges { limit: 10 }

# What are the main topic communities?
kg_communities {}

# Find shared context between two skills
kg_common { nodeA: "codex-delegation", nodeB: "task-routing" }

# Create a new decision record
kg_create_node {
  title: "Decision — Pricing Model 2026-Q3",
  directory: "memory",
  content: "## Context\n...",
  frontmatter: { tags: ["decision"], date: "2026-04-06" }
}

# Re-index after adding many new files
kg_index { resolution: 1.0 }

Re-indexing

Run kg_index via the MCP tool whenever:

• A significant batch of new files was added to the vault
• Wiki-links were restructured across many files
• Community assignments look stale

Or from CLI:

cd /c/Users/User/tools/knowledge-graph
KG_VAULT_PATH="C:/Users/User/Business-Empire-Agent" KG_DATA_DIR="C:/Users/User/tools/knowledge-graph/data" npx tsx src/cli/index.ts index

Notes

• Semantic search lazy-loads a transformer model on first use — expect ~5s delay for the first semantic call per session. Subsequent calls are fast.
• Full-text search (fulltext: true) is instant — use it when the query is a known term, not a concept.
• Node names are fuzzy-matched. Use a distinctive substring if the full path is long (e.g., "ACTIVE_TASKS" not "memory/ACTIVE_TASKS.md").
• kg_index resolution: Lower values (0.5) produce fewer, larger communities. Higher values (2.0) produce more granular clusters. Default 1.0 works well for this vault size.
• Index is stored at C:\Users\User\tools\knowledge-graph\data\kg.db — persists across sessions.

Obsidian Links

• [[brain/CAPABILITIES]] | [[brain/STATE]] | [[skills/memory-management/SKILL]]