Neuronest

Your AI's private doorway into your Second Brain.

NeuroNest MCP is a robust, local-first, minimal-dependency Model Context Protocol (MCP) server that securely connects any MCP-compliant AI agent (Claude, Cursor, etc.) to your Markdown-based knowledge base (Obsidian, Logseq, or plain folders).

Features

• Local-First & Private: Zero cloud dependencies, zero tracking. All processing happens on your machine.
• Multi-Vault Support: Connect multiple folders or Obsidian vaults simultaneously.
• Smart Snippet Extraction: Doesn't dump entire files into context. Intelligently extracts relevant paragraphs and preserves Markdown heading hierarchy.
• Semantic Search: Lightweight local embeddings (using sentence-transformers) for hybrid keyword + semantic search.
• Daily Digest Protocol: An executive assistant primitive that scans for tasks and highlights (#idea, #task, etc.) modified in the last 24 hours.
• Knowledge Graph Extraction: Understands [[wikilinks]] and #tags to navigate related concepts without a graph database.
• Change Awareness: Gives AI temporal context on what you modified recently.
• Token Budgeting: Actively manages LLM context window size using standard token estimation (tiktoken).

Quick Start

1. Install dependencies:

   pip install fastmcp pydantic tiktoken sentence-transformers
   

2. Configure your Vault Path: Set the environment variable or let the server default to the current directory.

   export NEURONEST_VAULT_PATH="/path/to/your/obsidian/vault"
   

3. Run the Server:

   python neuronest/server.py
   

Development

NeuroNest uses fastmcp to expose tools to Claude or Cursor. See GUIDE.md for a complete breakdown of tools and architecture.