Groot DataKG MCP Server - A Data Librarian's Desk

Model Context Protocol (MCP) server for enabling data extraction and context retrival for agents, stored in a knowledge graph using Amazon Neptune or FalkorDB. This is an experimental data modeller tool enabling semantic search to assit Agent to be a data librarian.

Architecture Overview

graph TB
    subgraph "MCP Clients"
        A[Claude Desktop]
        B[Kiro IDE]
        C[Other Agent Clients]
    end
    
    subgraph "Graph Memory MCP Server"
        D[MCP Protocol Handler]
        E[Memory Manager]
        F[Vector Search Engine]
        G[Query Engine]
        H[Entity/Relation Manager]
    end
    
    subgraph "Backend Databases"
        I[Amazon Neptune<br/>Cloud Scale]
        J[FalkorDB<br/>Redis-based]
    end
    
    subgraph "Core Features"
        K[Semantic Search<br/>AI Embeddings]
        L[Knowledge Graph<br/>Entities & Relations]
        M[Persistent Memory<br/>Agent Context]
        N[Graph Traversal<br/>Complex Queries]
    end
    
    A -.->|MCP Protocol| D
    B -.->|MCP Protocol| D
    C -.->|MCP Protocol| D
    
    D --> E
    E --> F
    E --> G
    E --> H
    
    F --> K
    G --> L
    H --> M
    G --> N
    
    E -->|OpenCypher/Gremlin| I
    E -->|OpenCypher| J
    
    style D fill:#e1f5fe
    style E fill:#f3e5f5
    style I fill:#fff3e0
    style J fill:#e8f5e8

Key Components

• MCP Protocol Handler: Manages communication with MCP clients using standard protocol

• Memory Manager: Core orchestration layer handling memory operations and backend selection

• Vector Search Engine: AI-powered semantic search using sentence transformers (384-dim embeddings)

• Query Engine: Executes OpenCypher and Gremlin queries against graph databases

• Entity/Relation Manager: Handles CRUD operations for graph entities and relationships

Data Flow

1. MCP Clients (Claude Desktop, Kiro IDE, etc.) connect via standard MCP protocol

2. Memory operations are processed through the Memory Manager

3. Semantic search uses AI embeddings for conceptual entity matching

4. Graph queries are executed against Neptune (cloud) or FalkorDB (local/Redis)

5. Results are returned through the MCP protocol to clients

Related MCP server: Graphiti MCP Server

Prerequisites

1. Install uv from Astral or the GitHub README

2. Install Python using uv python install 3.12

Installation

Local Development

# Clone the repository
git clone <repository-url>
cd ws-memory-mcp-server

# Install dependencies
uv sync

# Run with Neptune backend (full access mode)
uv run ws-memory-mcp-server --backend neptune --endpoint "neptune-db://your-cluster-endpoint"

# Run with FalkorDB backend (full access mode)
uv run ws-memory-mcp-server --backend falkordb --falkor-host localhost --falkor-port 6379

# Run in read-only mode
uv run ws-memory-mcp-server --backend falkordb --falkor-host localhost --falkor-port 6379 --mode read

# Run with SSE transport and custom logging
uv run ws-memory-mcp-server --backend falkordb --falkor-host localhost --falkor-port 6379 --sse --log-level DEBUG --log-file ./logs/mcp-server.log

MCP Client Configuration

Below are examples of how to configure your MCP client for different backends:

Neptune Backend

{
  "mcpServers": {
    "Neptune Memory": {
      "command": "uvx",
      "args": [
        "ws-memory-mcp-server",
        "--backend", "neptune",
        "--endpoint", "neptune-db://your-cluster-endpoint"
       ],
      "env": {
        "FASTMCP_LOG_LEVEL": "INFO"
      }
    }
  }
}

FalkorDB Backend

{
  "mcpServers": {
    "FalkorDB Memory": {
      "command": "uvx",
      "args": [
        "ws-memory-mcp-server",
        "--backend", "falkordb",
        "--falkor-host", "localhost",
        "--falkor-port", "6379",
        "--graph-name", "memory"
       ],
      "env": {
        "FASTMCP_LOG_LEVEL": "INFO"
      }
    }
  }
}

Backend Configuration

Neptune

When specifying the Neptune Endpoint the following formats are expected:

For Neptune Database: neptune-db://<Cluster Endpoint>

For Neptune Analytics: neptune-graph://<graph identifier>

FalkorDB

FalkorDB can be run locally using Docker:

# Run FalkorDB instance
docker run --rm -p 6379:6379 falkordb/falkordb

Or use FalkorDB Cloud for a managed instance.

Features

The MCP Server provides an agentic memory capability stored as a knowledge graph with support for:

• Multiple Backends: Choose between Amazon Neptune (cloud-scale) or FalkorDB (lightweight, Redis-based)

• Persistent Memory: Store agent memories as entities and relationships

• Knowledge Graph: Leverage graph database capabilities for complex relationship modeling

• MCP Integration: Standard MCP server interface for seamless integration with MCP clients

• Search & Query: Full-text search and graph traversal capabilities

• Observation Tracking: Store and manage observations about entities over time

Command Line Options

Common Options

• --backend: Database backend (neptune or falkordb, default: neptune)

• --mode: Server mode (read, write, or full, default: full)

  • read: Read-only access to the knowledge graph

  • write: Write-only access for creating and modifying data

  • full: Complete access with all read and write operations

• --sse: Enable SSE transport

• --port: Server port (default: 8888)

• --log-level: Logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL, default: INFO)

• --log-file: Path to log file for persistent logging

Neptune-Specific Options

• --endpoint: Neptune endpoint (required for Neptune)

• --use-https: Use HTTPS for Neptune connection (default: True)

• --no-https: Disable HTTPS for Neptune connection

FalkorDB-Specific Options

• --falkor-host: FalkorDB host (default: localhost)

• --falkor-port: FalkorDB port (default: 6379)

• --falkor-password: FalkorDB password

• --falkor-ssl: Use SSL for FalkorDB connection

• --graph-name: Graph name for FalkorDB (default: memory)