mem0-cathedral-mcp

1818TusculumSt/mem0-cathedral-mcp

3.1

If you are the rightful owner of mem0-cathedral-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

Cathedral MCP Server (Python Edition) provides intelligent memory management for Claude Desktop, enhancing memory quality and search capabilities.

Mem0 Cathedral MCP Server v12.2.0 (The Async Ascension)

Intelligent memory management for Claude Desktop - AI extraction, auto-recall, truly silent async operations, and advanced features.

What's New in v12.2.0? ⚡

Async Mode is now enabled by default for truly silent, lightning-fast memory operations:

  • ⚡ Instant Response: Memory saves return immediately - no waiting for extraction
  • 🤫 Truly Silent: Minimal latency means LLMs won't mention "saving to memory"
  • 🚀 Better Performance: Background processing improves scalability and response times
  • 📦 Mem0 v1.0.0+: Updated to latest Mem0 SDK with async-first architecture

How It Works

When you save a memory, the API now queues it for background processing instead of waiting. This means:

  • ✅ Faster tool responses (< 100ms instead of 2-3 seconds)
  • ✅ No verbose "I've saved that" messages cluttering chat
  • ✅ Better scalability for high-frequency memory operations

Features from v12.1.0 🚀

Complete rewrite bringing feature parity with the API version:

  • 🤖 AI Extraction Mode: Pass conversation messages, Mem0's AI extracts memories automatically
  • 🧠 Intelligent Auto-Recall: Proactive context retrieval with keyword reranking
  • 🤫 Silent Operations: Minimal responses ({success: true/false}) - no chat clutter
  • 📂 Custom Categories: 12 default categories for organized memories
  • 🕸️ Graph Memory: Entity relationship tracking for better context
  • 🤝 Multi-Agent Support: Track memories by agent_id and run_id
  • ⚡ Hybrid Search: Semantic (Mem0) + Lexical (keyword) matching
  • 🎯 Extraction Instructions: Guide AI extraction with custom prompts

Features

Core Operations

  • ✅ add-memory - AI extraction OR manual save with quality checks
  • ✅ get-context - Intelligent auto-recall with keyword reranking (NEW!)
  • ✅ search-memories - Enhanced semantic search with categories & graph
  • ✅ get-all-memories - Retrieve all memories for context loading
  • ✅ update-memory - Modify existing memories
  • ✅ delete-memory - Remove specific memories
  • ✅ consolidate-memories - Find and merge similar memories

Intelligence Features

  • AI Extraction: Mem0's LLM automatically extracts memories from conversations
  • Quality Gating: Minimum length, word count, and content validation
  • Duplicate Detection: Semantic similarity checking before save
  • Keyword Reranking: Hybrid semantic + lexical search (15% boost per keyword match)
  • Context Enrichment: Auto-adds timestamps and clarifying prefixes
  • Consolidation: Identifies redundant memories for cleanup
  • Silent Operations: No verbose responses cluttering your chat

Installation

Requirements

Setup

  1. Clone to your MCP tools location:

    git clone <repo-url> C:\mcptools\mem0-cathedral-mcp
cd C:\mcptools\mem0-cathedral-mcp

  2. Install Python dependencies:

    # Using pip
pip install -r requirements.txt

# Or using uv (recommended)
uv pip install -r requirements.txt

  3. Update Claude Desktop config (%APPDATA%\Claude\claude_desktop_config.json):

    {
  "mcpServers": {
    "mem0": {
      "command": "python",
      "args": ["C:\\mcptools\\mem0-cathedral-mcp\\server.py"],
      "env": {
        "MEM0_API_KEY": "your_mem0_api_key_here"
      }
    }
  }
}

  4. Restart Claude Desktop

Configuration

Environment Variables

  • MEM0_API_KEY (required): Your Mem0 API key
  • Default user ID: el-jefe-principal

Quality Thresholds (in server.py)

Customize these constants to adjust filtering:

MIN_MEMORY_LENGTH = 20      # Minimum characters
MIN_WORD_COUNT = 4          # Minimum words
SIMILARITY_THRESHOLD = 0.85 # Deduplication threshold (0.0-1.0)

Usage Examples

Adding Memories - AI Extraction Mode (RECOMMENDED)

Let Mem0's AI extract automatically:

{
  "messages": [
    {"role": "user", "content": "I love pizza but hate pineapple on it"},
    {"role": "assistant", "content": "I'll remember your pizza preferences!"}
  ]
}

Mem0 extracts: "User loves pizza" + "User dislikes pineapple on pizza"

Response: {"success": true} (silent operation)

Adding Memories - Legacy Mode

Manual extraction with quality filtering:

{
  "content": "User prefers TypeScript over JavaScript for type safety"
}

Response: {"success": true} or {"success": false} (rejections are silent)

Auto-Recall Context

Get relevant memories for current conversation:

{
  "currentMessage": "What should I order for dinner?",
  "recentMessages": [
    {"role": "user", "content": "I'm hungry"},
    {"role": "assistant", "content": "What sounds good?"}
  ]
}

Response:

{
  "context": "## User Context\n\n### Food Preferences\n- User loves pizza\n- User dislikes pineapple on pizza\n\n",
  "count": 2
}

Searching Memories

Enhanced search with categories and graph:

{
  "query": "programming languages",
  "categories": ["technical", "preferences"],
  "enableGraph": true,
  "limit": 10
}

Use get-context instead for auto-recall - it includes keyword reranking!

Consolidating Memories

User: Clean up my memories
Claude: [Calls consolidate-memories with dryRun: true]

Response: Found 5 similar memory pairs:
  1. "User prefers Python" ↔ "User likes Python for data science" (similarity: 0.82)
  2. "Working on MCP project" ↔ "Building Mem0 MCP server" (similarity: 0.76)
  ...

Architecture

  • Language: Python 3.10+
  • MCP SDK: mcp (official Python SDK)
  • Mem0 Client: mem0ai (official Python SDK)
  • Protocol: stdio (standard input/output)
  • Async: Full async/await support

Version Comparison

FeatureNode.js (v1.0)Python (v2.0)Python (v12.1.0)
Quality Filtering❌✅✅
Deduplication❌✅✅
Context Enrichment❌✅✅
Memory Consolidation❌✅✅
AI Extraction❌❌✅
Auto-Recall❌❌✅
Silent Operations❌❌✅
Custom Categories❌❌✅
Graph Memory❌❌✅
Multi-Agent Support❌❌✅
Keyword Reranking❌❌✅

API Quality Features

Memory Quality Assessment

Before saving, each memory is scored:

  • ✅ Good: Length ≥20 chars, ≥4 words, has context indicators
  • ⚠️ Warning: Very long (>500 chars) - may need summarization
  • ❌ Reject: Too short, acknowledgments only, no context

Context Indicators

Memories containing these terms get bonus points:

  • Preferences: "prefer", "like", "love", "hate", "always", "never"
  • Technical: "project", "tool", "language", "technology"
  • Personal: "name is", "location", "timezone", "schedule"
  • Goals: "goal", "objective", "plan", "want to", "need to"

Troubleshooting

Server not starting

  • Check Python version: python --version (need 3.10+)
  • Verify MEM0_API_KEY is set in config
  • Check logs: %APPDATA%\Claude\logs\mcp-server-mem0.log

Dependencies not found

# Reinstall dependencies
pip install --upgrade -r requirements.txt

Too many memories rejected

Lower quality thresholds in :

MIN_MEMORY_LENGTH = 10   # Was 20
MIN_WORD_COUNT = 2       # Was 4

Search not triggering

Tool description has been optimized, but you can further customize at

Development Workflow

Dev/Prod Split

  • Dev: C:\Users\jonat\OneDrive\Coding Projects\mem0-cathedral-mcp\
  • Prod: C:\mcptools\mem0-cathedral-mcp\

Testing Locally

# Set API key
set MEM0_API_KEY=your_key_here

# Run server directly
python server.py

# Test with MCP inspector
mcp dev server.py

Deploying to Production

cd C:\mcptools\mem0-cathedral-mcp
git pull
# Restart Claude Desktop

Version History

  • 12.2.0 - The Async Ascension: Async mode enabled by default, instant responses, upgraded to Mem0 v1.0.0+
  • 12.1.0 - The Silent Oracle: AI extraction, auto-recall, silent operations, full feature parity with API version
  • 2.0.0 - Python rewrite with intelligent filtering, deduplication, consolidation
  • 1.0.0 - Original Node.js cathedral implementation

Upgrade Guide

From v12.1.0 to v12.2.0

What Changed:

  • Upgraded to Mem0 v1.0.0+ (from v0.1.x)
  • Async mode (async_mode=True) is now explicitly enabled for memory saves
  • Memory operations return instantly (queued for background processing)
  • Even faster, more silent operation

Migration Steps:

  1. Pull latest code: git pull
  2. Update dependencies: pip install --upgrade -r requirements.txt
  3. Restart Claude Desktop
  4. Enjoy instant memory saves!

Backward Compatible: No breaking changes, just better performance!

From v2.0.0 to v12.1.0

What Changed:

  • Tool responses are now SILENT by default ({success: true/false})
  • New get-context tool for intelligent auto-recall
  • add-memory now supports AI extraction via messages parameter
  • Enhanced search with categories, graph memory, multi-agent support

Migration Steps:

  1. Pull latest code: git pull
  2. Restart Claude Desktop
  3. Start using get-context for proactive memory recall!
  4. Optionally use messages mode for better extraction

Backward Compatible: Old add-memory with content still works!

Credits

Originally based on mem0-cathedral-api for Open WebUI.

v12.1.0 upgrade brings full feature parity between API and MCP versions. 🧠

Default Memory Categories

The following 12 categories are used when AI extraction is enabled:

  • personal_information - Name, location, age, family, background
  • preferences - Likes, dislikes, favorites, personal tastes
  • work - Career, projects, professional information
  • food_preferences - Food likes, dislikes, dietary restrictions
  • technical - Tech stack, tools, programming languages
  • goals - Objectives, plans, aspirations, future intentions
  • health - Health conditions, fitness routines, wellness
  • hobbies - Interests, activities, pastimes
  • relationships - Friends, family, colleagues, connections
  • location - Places lived, traveled, or frequently visited
  • schedule - Routines, availability, time preferences
  • communication - Preferred communication styles and channels

You can override these by passing customCategories to add-memory.