MegaMem — Obsidian × Knowledge Graph × MCP

Temporal memory for AI assistants, powered by your Private vault

MegaMem is an Obsidian plugin that syncs your notes into a temporal knowledge graph (powered by Graphiti) and exposes it to AI assistants through the Model Context Protocol (MCP). Claude, and any other MCP-compatible client, can read, search, and write to your vault — and remember things across conversations.

---

🌟 Key Features

🧠 Temporal Knowledge Graph

Sync your Obsidian notes to a graph database. Entities become nodes, relationships are extracted by AI, and every fact is timestamped so the graph evolves as your knowledge does — without losing history.

🔍 Auto Schema Discovery

The plugin scans your vault's frontmatter and automatically infers entity types, property types, and relationships. No manual schema definition required. Write your notes, the schema emerges.

🌐 Multi-Database & Multi-Vault (new in v1.5.0)

Configure multiple named graph databases simultaneously — each with its own connection, type (Neo4j/FalkorDB), and embedding model. A masterVault runs the MCP server and manages all databases across all registered vaults. Tell Claude which database to query, or let it discover available databases with list_databases.

🤖 21 MCP Tools for AI Assistants

A full MCP server (11 graph tools + 10 vault file tools) gives Claude — or any MCP client — direct, structured access to your knowledge. Search memories, add episodes, read and write notes, explore folders, all from your AI conversation.

🏗️ Custom Ontology Manager

Define your own entity types, edge types, and property descriptions. Generate Pydantic models with one click. The AI understands your custom data structures during extraction.

⚡ Obsidian CLI Integration

All 10 file operation tools run through the native Obsidian CLI (v1.12+) — stateless subprocess calls with no persistent WebSocket, no connection race conditions, no heartbeat. Multi-vault support via a single vault_id parameter. Non-markdown files (.pdf, .png, .csv, .base, etc.) are now fully supported across all file tools.

🌐 Streamable HTTP MCP Transport (new in v1.4.0)

Connect Roo Code, Cursor, Claude Code, VS Code, NemoClaw, and any HTTP-capable MCP client directly to MegaMem — no Claude Desktop required. Enable the opt-in Streamable HTTP server (MCP spec 2025-03-26) in Plugin Settings → Servers. Token Profiles let you create scoped bearer tokens per client, each with a one-click Copy Config for all 6 supported clients. Remote access via Tailscale is supported via a per-profile Endpoint URL field.

🔄 Intelligent Sync

Auto-sync on interval, sync on demand, or trigger from MCP. Filter by folder inclusion/exclusion. Choose "new only" or "new + updated". Sync state is tracked in a local SQLite database (sync.db) — per-note, per-database, with content hashing to skip unchanged notes and full analytics logging (token counts, entity counts, sync duration).

📊 Sync Analytics Dashboard (new in v1.5.5)

The unified MegaMem panel (brain icon, copper accent) combines Ontology, Sync, and Analytics in a single tabbed sidebar. The Analytics tab provides a comprehensive dashboard: animated summary cards (Synced Notes, Entities Extracted, Edges Created, Estimated Cost), a Sync Timeline chart, Token Usage by Model (Big / Small / Both Model views), a sortable Model Performance table with Provider column, a Synced Notes accordion with per-session detail (In/Out/Total tokens, duration, small model, error message), and sync health indicators. Pricing data is seeded from your Model Library and updated after every provider fetch.

✦ MegaMem Pro (new)

A dedicated Pro tab in plugin settings for licensed Stewards. Validate your API key, check and install content packages (vault templates, ontology packs), and access upcoming hosted services — all in-plugin. Free plan users keep all MCP tools; Pro is content delivery and future hosted features.

---

🚀 BETA Launch

MegaMem is stable in daily production use and currently in public beta. We're actively seeking testers to stress-test the system across diverse environments — different vaults, databases, LLM providers, and operating systems. If you find something, open an issue. If it works great, tell someone.

• Install now via BRAT → C-Bjorn/megamem-mcp
• Python components install automatically on first launch — no manual downloads
• Windows (Neo4j) is most battle-tested; macOS works great; Linux support is there but less tested

---

🛠️ MCP Tools Reference

All 21 tools are available to Claude Desktop and any MCP-compatible client.

Graph Operations (11)

Tool	Description
add_memory	Add an episode/memory to the knowledge graph (database_id optional)
add_conversation_memory	Store a conversation as a structured memory episode
search_memory_nodes	Semantic search for entity nodes in the graph (database_id optional)
search_memory_facts	Search for relationships and facts between entities (database_id optional)
get_episodes	Retrieve the most recent N episodes from a group
get_entity_edge	Get relationships for a specific entity by name
delete_entity_edge	Remove a specific relationship edge by UUID
delete_episode	Remove a specific episode by ID
list_group_ids	List all group IDs (namespaces) in the vault
list_databases	List all configured database targets — use before routing with database_id
clear_graph	Clear the entire memory graph (use with caution)

Obsidian File Operations (10) — via Obsidian CLI

Non-markdown files (.pdf, .png, .csv, .base, etc.) are supported across all file tools — the _auto_md fix skips .md auto-append when the path already has a recognized extension.

Tool	Description
search_obsidian_notes	Search vault notes by filename and/or content
read_obsidian_note	Read a note's full content (with optional line map for editing)
update_obsidian_note	Update a note — 5 modes: full file, frontmatter, append, range, editor
create_obsidian_note	Create a new note at a specified path
list_obsidian_vaults	List all registered Obsidian vaults
explore_vault_folders	Explore vault folder structure (tree/flat/paths output)
create_note_with_template	Create a note using a Templater template with intelligent routing
manage_obsidian_folders	Create, rename, or delete vault folders
manage_obsidian_notes	Delete or rename/move notes (cross-folder moves supported)
manage_obsidian_base	Manage Obsidian Bases .base files — operations: list, views, query, create

Full parameter reference: docs/mcp-commands.md — Updated for v1.5 with database_id routing on all graph tools.

---

📋 How It Works

1. Schema Discovery

The plugin scans your vault frontmatter and infers entity types and properties:

---
type: Person
name: "Jane Smith"
role: "Researcher"
organization: "TechCorp"
tags: ["AI", "Knowledge Graphs"]
---

2. Pydantic Model Generation

Your frontmatter patterns become typed Graphiti extraction models:

class Person(BaseNode):
    name: Optional[str] = Field(None, description="Person's full name")
    role: Optional[str] = Field(None, description="Current role or title")
    organization: Optional[str] = Field(None, description="Current employer")

3. Temporal Graph Sync

Notes are processed through Graphiti's extraction pipeline:

• Entities become graph nodes with versioned properties
• Relationships are extracted from note content via LLM
• Sync state tracked in SQLite (sync.db) per-note/per-DB with content hashing — unchanged notes are skipped automatically
• AI assistants query this structured graph via MCP

---

🚀 Getting Started

Prerequisites

Component	Minimum	Notes
Obsidian	1.12.4+ (installer)	Must run the full installer — in-app updates don't enable CLI
Python	3.11+	Managed automatically via UV
Database	Neo4j 5+ or FalkorDB 1+	Neo4j Desktop is easiest for local use

⚠️ Obsidian installer required: Download and run the full installer from obsidian.md/download. In-app auto-update does not update the CLI binary. Then go to Settings → General → Command line interface → Register.

Install via BRAT (recommended)

1. Install BRAT from Community Plugins
2. BRAT → Add Beta Plugin → paste C-Bjorn/megamem-mcp
3. A dialog appears after ~3 seconds: click "Install / Update Components"
   • Downloads and installs graphiti_bridge + mcp-server automatically
4. Plugin is ready

7-Step Setup

Step 0 — Register Obsidian CLI (required for file tools) Run the Obsidian 1.12.4+ installer → Settings → General → CLI → Register → restart terminal → verify with obsidian version

Step 1 — Set up a graph database Neo4j Desktop (recommended) or FalkorDB via Docker. See Database Setup Guide.

Step 2 — Install Python dependencies Plugin Settings → Python Environment → "Install Dependencies" (uses UV by default — works on macOS/Windows/Linux).

Step 3 — Configure LLM provider Plugin Settings → API Keys → enter your key. Click "Load Defaults" to populate recommended models.

Step 4 — Connect the database Plugin Settings → Database Configuration → enter your connection details → "Test Connection" → "Initialize Schema".

Step 5 — Connect your MCP client

• Claude Desktop (stdio): Plugin Settings → Servers → STDIO MCP Server → "Generate Config" → paste into claude_desktop_config.json → restart Claude Desktop.
• HTTP clients (Roo Code, Cursor, Claude Code, VS Code, NemoClaw): Plugin Settings → Servers → Streamable HTTP Access → enable toggle → expand the Admin token profile → click the Copy Config button for your client. No Claude Desktop needed.

Step 6 — (Optional) Configure sync Advanced sync settings (auto-sync toggle, sync interval, included folders) are in Plugin Settings → Advanced Settings tab → Auto-Sync. Excluded folders and per-note extraction settings are in General → Sync Configuration.

Step 7 — Start syncing Three ways to sync:

• Single note: click the sync icon (top-right of any note window) to sync the current note
• Bulk sync / Analytics: click the 🧠 brain icon in the ribbon to open the unified MegaMem panel → switch to the Sync tab for bulk operations
• Command palette: MegaMem: Open Sync / MegaMem: Open Analytics / MegaMem: Open Ontology

⚠️ Notes must have a type property in their frontmatter to use bulk Sync (when Custom Ontology is enabled).

---

🎯 Use Cases

Personal Knowledge Management

Build a graph of the people, projects, ideas, and relationships in your life. Ask Claude "what do I know about X?" and get answers drawn from your actual notes, with full temporal context. It works as a personal CRM, project tracker, and second brain — all from your existing vault without restructuring anything.

AI-Assisted Note-Taking

With 9 native Obsidian file tools, Claude can create, search, and update notes directly in your vault mid-conversation. Draft a meeting note, file it in the right folder, sync it to the graph — all in one step, without leaving the chat.

Also great for: research & academia (literature graphs, citation tracking), business intelligence (competitor research, knowledge bases), creative projects (world-building, character arcs), and multi-team knowledge management with namespaced vaults.

---

📚 Documentation

Getting Started

• Introduction — Overview and core concepts
• Quick Start Guide — Up and running in minutes
• Plugin Settings Reference — All configuration options

Guides

• Database Setup — Neo4j and FalkorDB configuration
• Claude Desktop Integration — MCP config and connection
• Ontology Manager — Custom entity types and schemas
• Sync Manager — Sync behavior and controls

Reference

• MCP Commands — All 19 tools with full parameter reference
• FAQ — Common questions answered

---

🤝 Community & Support

• GitHub Issues — Bug reports and feature requests
• GitHub Discussions — Questions, use cases, show & tell
• Submit & Fund Features — Vote on and fund roadmap items
• Contributing Guide — How to contribute code or docs

---

🔮 Roadmap

Shipped ✅

• Unified MegaMem Panel & Sync Analytics Dashboard — brain icon ribbon, tabbed panel (Ontology | Sync | Analytics), animated summary cards, Timeline chart, Token Usage by Model (Big/Small/Both), sortable Model Performance table with Provider column, Synced Notes accordion, Sync Health indicators, bundled model pricing (v1.5.5)
• SQLite Sync Registry — sync.db replaces sync.json; per-note/per-DB content hashing (skip unchanged), full analytics logging (token counts, entity/edge counts, duration, model used), episode UUID chain via SQLite (no more mm_uid frontmatter writes) (v1.5.5)
• Multi-database support — multiple named Neo4j/FalkorDB targets, per-DB embedding config, sync dropdown per note (v1.5)
• Multi-vault architecture — masterVault control panel, childVault registration, MCP database_id routing, list_databases tool (v1.5)
• Temporal knowledge graph sync (Graphiti + Neo4j/FalkorDB)
• 21 MCP tools — 11 graph operations + 10 Obsidian file tools (including manage_obsidian_base for .base files)
• Obsidian CLI integration (stateless, multi-vault, no WebSocket)
• Auto schema discovery from vault frontmatter
• Custom ontology manager with Pydantic model generation
• Model Library — live model fetching from 8 LLM providers; pricing synced to analytics DB for cost tracking
• Constrained ontology generation (edge type cap + deduplication)
• Ontology file separation (ontology.json split from data.json)
• Auto-update system — Python components auto-install and self-update
• MegaMem Pro tab — license validation + content package delivery

In Progress / Near-term 🔄

• Obsidian Community Plugins submission
• Graph visualization within Obsidian
• Advanced query builder

Future 🌱

• Built-in LLM chat interface inside Obsidian
• Cloud-hosted graph option (no local database required)
• Mobile support
• Additional database backends (Kuzu, Amazon Neptune)

---

💡 Philosophy

MegaMem embraces progressive formalization: start with plain notes, let patterns emerge naturally, then add structure as your knowledge matures. The graph enriches your vault — it doesn't replace the way you write.

Gate services, not tools. All 20 MCP tools are free. Forever.

---

🙏 Built With

• Graphiti by Zep — temporal knowledge graph engine
• Obsidian API — plugin foundation
• Model Context Protocol by Anthropic — AI tool interface
• JSZip — in-plugin zip extraction
• TypeScript + Svelte

📄 License

MIT — see LICENSE.txt

---

Transform your notes into knowledge. Give your AI a memory worth having.

Built by Casey Bjørn @ ENDOGON