⚡ MCP KNOWLEDGE GRAPH ⚡

How Claude Remembers Across Conversations

TRIPLE-MODE HTML ARCHITECTURE

👁️ HUMAN VISUAL: Interactive Schematic
🤖 AI MEMORY: Pattern Recognition
🔍 SEO/SEARCH: Schema.org Indexing

THE PROBLEM IT SOLVES

❌ WITHOUT MCP MEMORY

  • Every conversation starts fresh
  • Claude has amnesia between sessions
  • You repeat context constantly
  • Project continuity breaks
  • Function names forgotten
  • Architecture decisions lost

✅ WITH MCP MEMORY

  • Persistent memory across all conversations
  • Claude remembers people, projects, facts
  • Automatic context retrieval
  • Project state preserved
  • Code patterns remembered
  • True continuity achieved

HOW IT ACTUALLY WORKS

1
Claude Desktop Boots
Reads claude_desktop_config.json and launches MCP servers as Node.js background processes
⬇
2
MCP Server Runs
Command: npx -y mcp-knowledge-graph --memory-path /path/to/memory.jsonl
This creates a local API that Claude can call
⬇
3
Conversation Starts
Custom instructions trigger: Claude says "Remembering..." and calls read_graph()
⬇
4
Memory Retrieval
MCP server reads memory.jsonl and returns entities/relations to Claude
⬇
5
You Share Information
"I'm working on Eureka Gain Score - it's a React app with PostgreSQL"
⬇
6
Claude Stores Memory
Calls create_entities() and add_observations() to persist information
⬇
7
Data Persists
MCP server writes to memory.jsonl - file survives forever
⬇
8
Next Conversation
New Claude instance reads the same memory.jsonl and remembers everything!

TECHNICAL ARCHITECTURE

MCP (Model Context Protocol)

Anthropic's standard for connecting external tools to Claude

Node.js Process

Background server running locally on your machine

Knowledge Graph

Entities + Relations + Observations = Structured memory

JSONL Storage

Each line = one entity/relation, easy to append/read

KNOWLEDGE GRAPH STRUCTURE

ENTITY (Person)

{ "name": "Russell_Wright", "entityType": "person", "observations": [ "Founder of Super Intelligent AI SEO", "25+ years ML experience", "Located in Arizona" ] }

ENTITY (Project)

{ "name": "Eureka_Gain_Score", "entityType": "project", "observations": [ "React frontend", "PostgreSQL backend", "Active development" ] }

RELATION (Connection)

{ "from": "Russell_Wright", "to": "Eureka_Gain_Score", "relationType": "works_on" }

🎯 THE BREAKTHROUGH

This isn't magic - it's structured tool calling that writes to a persistent file.


Every time Claude calls create_entities() or add_observations(), the MCP server appends a line to memory.jsonl. That file lives on your disk forever.


Next conversation? New Claude instance reads that same file. Instant continuity.


It's not AI remembering - it's AI reading its own notes from last time. Brilliant in its simplicity.

INTEGRATION WITH LIVING CONTEXT ENGINE

📚 LIVING CONTEXT ENGINE

  • Curated markdown files
  • Big patterns & philosophy
  • Manual loading sequence
  • Human-readable insights
  • Constitutional foundation

🧠 MCP KNOWLEDGE GRAPH

  • Automatic JSONL database
  • Specific facts & entities
  • Zero maintenance needed
  • Machine-optimized storage
  • Daily operational memory
They're complementary, not competing.
Living Context Engine = Identity & Patterns
MCP Knowledge Graph = Facts & Relationships

HOW TO SET IT UP

1
Edit Config File
Add memory server to claude_desktop_config.json
{ "mcpServers": { "memory": { "command": "npx", "args": ["-y", "mcp-knowledge-graph", "--memory-path", "/path/to/memory.jsonl"] } } }
2
Add Custom Instructions
In claude.ai Profile settings, add the memory protocol instructions
3
Restart Claude Desktop
MCP servers load on startup - restart required for changes to take effect