Zettelkasten - MCP Server for Chatbot Applications

March 29, 2025 · View on GitHub

Zettelkasten is an MCP (Message Control Protocol) server designed for chatbot applications. This repository provides two implementations of an MCP server:

Stdio Transport
Server-Sent Events (SSE) Transport

Features

Hybrid Data Storage: Combines a vector database (Pinecone) and a graph database (Neo4j) for efficient storage and retrieval of cross-session chat context.
Multi-Protocol Support: Offers both Stdio and SSE transports to integrate seamlessly with various clients.
Scalability: Designed to handle large-scale chatbot interactions with contextual memory persistence.

Getting Started

Prerequisites

Ensure you have the following installed on your system:

Node.js (Latest LTS version recommended)
pnpm (Package manager)
Neo4j Database running locally or remotely
Pinecone Vector Database account
OpenAI API Key

Environment Variables

Create a .env file in the /apps/mcp-server directory and include the following environment variables:

PINECONE_API_KEY=
PINECONE_INDEX_NAME=
NEO4J_URI=
NEO4J_USER=
NEO4J_PASSWORD=
OPENAI_API_KEY=
EMBEDDING_MODEL=text-embedding-ada-002
VECTOR_DIMENSION=1536

Note: Do not share or commit your API keys.

Running Zettelkasten Locally

1. Stdio Transport

To set up and run the MCP server using stdio transport:

cd apps/mcp-server
pnpm install
pnpm build

If you are integrating with Claude, update claude_desktop_config.json:

"mcpServers": {
  "zettelkasten": {
    "command": "node",
    "args": [
      "<absolute-path-to>/apps/mcp-server/build/zettelkasten-local.js"
    ],
    "env": {
      "PINECONE_API_KEY": "",
      "PINECONE_INDEX_NAME": "",
      "NEO4J_URI": "",
      "NEO4J_USER": "",
      "NEO4J_PASSWORD": "",
      "OPENAI_API_KEY": ""
    }
  }
}

2. SSE Transport

To run the MCP server using SSE transport:

First, create a .env file in the /apps/mcp-server directory with the following environment variables:

PINECONE_API_KEY=
PINECONE_INDEX_NAME=
NEO4J_URI=
NEO4J_USER=
NEO4J_PASSWORD=
OPENAI_API_KEY=
EMBEDDING_MODEL=text-embedding-ada-002
VECTOR_DIMENSION=1536

Then, install dependencies and start the server:

cd apps/mcp-server
pnpm install
pnpm start

This starts the MCP server at http://localhost:3001. To use the server in an application like Cursor, configure it as follows:

"mcpServers": {
  "server-name": {
    "url": "http://localhost:3001/sse"
  }
}

Why Zettelkasten?

Zettelkasten leverages a hybrid approach by combining:

Vector databases (Pinecone) for semantic search and similarity-based context retrieval.
Graph databases (Neo4j) for relationship-based storage, ensuring chat context is efficiently structured and retrievable across sessions.

This combination provides fast, context-aware chatbot interactions, making it ideal for AI-driven conversational agents.