Give any AI agent a fully searchable, self-organizing knowledge base.
Brain is an MCP (Model Context Protocol) server that connects AI agents to your Obsidian vault. Full-text search, semantic embeddings, voice memo transcription, calendar sync — all exposed as tools your agent can call.
Runs as a sidecar to Hermes Agent or any MCP-compatible client. Powers my second-brain vault with 4,000+ notes.
AI agents have no long-term memory. They forget everything between sessions. Brain gives them persistent, searchable memory — your Obsidian vault becomes their knowledge base. They can read, search, write, and connect ideas across sessions.
┌──────────────────────────────────────────┐
│ AI Agent (Hermes / Claude / any MCP) │
└──────────────┬───────────────────────────┘
│ MCP (stdio)
┌──────────────▼───────────────────────────┐
│ Brain MCP Server (FastMCP) │
│ │
│ ┌─────────┐ ┌──────────┐ ┌───────────┐ │
│ │ Vault │ │ Calendar │ │ Whisper │ │
│ │ Ops │ │ Ops │ │ Server │ │
│ │ │ │ │ │ │ │
│ │ search │ │ CRUD │ │ local ≤4m │ │
│ │ embed │ │ sync │ │ Groq >4m │ │
│ │ ingest │ │ queue │ │ │ │
│ │ write │ │ │ │ │ │
│ └────┬────┘ └────┬─────┘ └─────┬──────┘ │
└───────┼───────────┼─────────────┼─────────┘
│ │ │
┌───────▼───┐ ┌─────▼──────┐ ┌───▼─────────┐
│ Obsidian │ │ Google │ │ Audio Files │
│ Vault │ │ Calendar │ │ (.mp3, .wav) │
└───────────┘ └────────────┘ └──────────────┘
| Feature | Tool | Description |
|---|---|---|
| Full-text search | search_vault |
Regex search across all markdown files |
| Semantic search | semantic_search |
ONNX embeddings, meaning-based retrieval |
| Audio ingestion | ingest_audio |
Voice memo → transcribed → vault (whisper + Groq) |
| Document ingestion | ingest_document |
PDF/text → auto-chunked → vault |
| Calendar | list_calendar_events |
Google Calendar CRUD + sync queue |
| Dashboard | update_dashboard |
Safe task management via dashboard.md |
| Server health | get_server_status |
CPU, RAM, disk, PM2 process health |
| Telegram bridge | send_telegram_question |
Dual-channel ask (CLI + Telegram buttons) |
# 1. Clone
git clone https://github.com/jacquesmyo/brain.git
cd brain
# 2. Install
pip install -e .
# 3. Point at your vault
export BRAIN_VAULT_PATH=~/vault
# 4. Start the server
python src/brain/server.pyThen configure your MCP client:
{
"mcpServers": {
"brain": {
"command": "python",
"args": ["/path/to/brain/src/brain/server.py"]
}
}
}Your agent now has a persistent knowledge base.
- Python 3.12+
- FastMCP — MCP server framework
- ONNX Runtime — embeddings (all-MiniLM-L6-v2, 80MB)
- whisper — local transcription (≤4 min audio)
- Groq API — remote transcription (>4 min audio)
- Google Calendar API — calendar sync
Brain runs 24/7 on a single Linux box alongside Hermes Agent. It handles:
- 4,000+ markdown notes in the vault
- Hourly calendar sync cron jobs
- Daily physics arxiv paper ingestion
- Voice memo capture → auto-filing
- Automated vault review at 04:00 daily
All CPU-only. No GPU needed.
- second-brain — The vault this server indexes
- moe-tools — AI research framework documented in the vault
- Hermes Agent — Primary MCP client
MIT — use it, fork it, ship it. Attribution appreciated.