KMS Unified MCP Memory System

Overview

This skill enables Claude Code to use persistent memory across all sessions through the KMS Unified MCP. Unlike traditional stateless interactions, you can now store and retrieve knowledge spanning multiple dimensions: personal preferences (Mem0), conceptual relationships (Neo4j), and structured data (MongoDB).

Core Principle: Multi-Dimensional Memory

Memory is naturally multi-faceted. Rich information benefits from storage across multiple systems simultaneously:

**Example:** "Client responds well to morning sessions and visualization techniques"

Available Tools

`unified_search` - Search Existing Knowledge

**Always search before storing** to avoid duplicates and build on existing context.

```json

{

"query": "client morning sessions visualization",

"filters": {

"contentType": ["memory", "insight", "relationship"],

"userId": "client_123"

},

"options": {

"includeRelationships": true,

"maxResults": 10

}

}

```

`unified_store` - Multi-Dimensional Storage

Store knowledge across systems based on its natural dimensions.

```json

{

"content": "Client breakthrough with morning visualization techniques",

"contentType": "insight",

"source": "coaching",

"userId": "client_123",

"relationships": [

{

"targetId": "morning-sessions",

"type": "ENHANCED_BY",

"strength": 0.9

}

]

}

```

Instructions for Claude Code

1. Develop Automatic Search Triggers

Make searching second nature when:

2. Integrated Problem-Solving Flow

**Old:** Question → Think → Answer

**New:** Question → **unified_search** → Think with context → Answer + **unified_store** new insights

3. Natural Storage Moments

Store when encountering:

4. Multi-Dimensional Thinking

When storing rich information, consider multiple aspects:

**Technical breakthrough:** "Fixed OAuth by updating JWKS endpoint"

**Client insight:** "Morning meditation helps client focus during difficult conversations"

5. Best Practices

**Search First, Store Smart:**

1. Always search before storing

2. Use search results to inform storage decisions

3. Build on existing knowledge, don't create isolated memories

**Natural Language Processing:**

**Context Awareness:**

**Multi-Dimensional Storage Example:**

```json

{

"content": "User prefers async communication over real-time meetings",

"contentType": "preference",

"source": "personal",

"userId": "user_123",

"metadata": {

"communication_style": "asynchronous",

"meeting_preference": "scheduled",

"context": "work_efficiency"

},

"relationships": [

{

"targetId": "communication-preferences",

"type": "INSTANCE_OF",

"strength": 0.9

},

{

"targetId": "productivity-patterns",

"type": "RELATES_TO",

"strength": 0.7

}

]

}

```

Datastore Strengths

SemTools Integration

The KMS MCP includes SemTools for semantic document intelligence:

**`parse`** - Convert PDFs, DOCX, PPTX to markdown (uses LlamaParse API)

**`search`** - Local semantic search with multilingual embeddings

**`workspace`** - Persistent caching for fast repeated searches

Common Patterns

```bash

Parse and search documents

parse document.pdf | xargs cat | search "error handling" --n-lines 30

Use workspaces for large collections

workspace use my-workspace

export SEMTOOLS_WORKSPACE=my-workspace

search "financial analysis" docs/*.pdf --n-lines 30 --top-k 10

Advanced pipelines

parse reports/*.pdf | xargs search "compliance" --n-lines 50 --max-distance 0.3

```

**Best Practices:**

Implementation Goal

Make memory integration automatic - like naturally breaking down complex problems or connecting related concepts. The unified MCP handles technical complexity while you focus on the conceptual richness of multi-dimensional memory.

**Positive Reinforcement:** The more you search and find useful previous knowledge, the more natural it becomes. When stored memories help future conversations, the value is clear.