---

name: qdrant-memory description: Use this skill for semantic search, long-term memory storage, and RAG (Retrieval Augmented Generation). Enables vector-based knowledge retrieval and persistent memory across sessions.

Qdrant Vector Memory

This skill enables semantic search and long-term memory using Qdrant vector database.

Problem Solved

Traditional keyword search:

• Exact match only
• Misses semantically similar content
• No context understanding

With Qdrant Vector Search:

• Semantic similarity matching
• Finds conceptually related information
• Understands context and meaning
• Persistent memory across sessions

When to Use

• Storing knowledge for later retrieval
• Semantic code search across codebase
• Building RAG (Retrieval Augmented Generation) systems
• Long-term memory for AI agents
• Finding similar documents/code
• Knowledge base management

Available Tools

1. qdrant-store

Stores text with vector embeddings for later retrieval.

CopyInput: { "text": "React hooks are functions...", "metadata": { "topic": "react" } }
Output: Stored with vector embedding

2. qdrant-find

Finds semantically similar content.

CopyInput: { "query": "how to manage state in React" }
Output: Related documents ranked by similarity

3. qdrant-delete

Removes stored memories by ID or filter.

CopyInput: { "filter": { "topic": "outdated" } }
Output: Deleted matching entries

4. qdrant-list-collections

Lists all available collections.

CopyOutput: Collection names and stats

Example Usage

Store Knowledge

CopyUser: このReactパターンを覚えておいて

AI: [Calls qdrant-store]
    [Embeds content with sentence-transformers]
    [Stores in taisun_memory collection]

→ 永続的に保存され、後で検索可能

Semantic Search

CopyUser: 以前話した状態管理のパターンは？

AI: [Calls qdrant-find with semantic query]
    [Returns top-k similar documents]
    [Provides context from stored memories]

RAG for Code Generation

CopyUser: 前回実装したAPIパターンを参考に新しいエンドポイント作って

AI: [Searches for similar API implementations]
    [Retrieves relevant code patterns]
    [Generates new code based on retrieved context]

Architecture

Copy┌─────────────────────────────────────────────────────┐
│                    TAISUN Agent                     │
└─────────────────────┬───────────────────────────────┘
                      │
                      ▼
┌─────────────────────────────────────────────────────┐
│              Qdrant MCP Server                      │
│  ┌───────────────┐  ┌───────────────────────────┐  │
│  │ Embedding     │  │ Vector Store              │  │
│  │ (MiniLM-L6)   │──│ (taisun_memory collection)│  │
│  └───────────────┘  └───────────────────────────┘  │
└─────────────────────┬───────────────────────────────┘
                      │
                      ▼
┌─────────────────────────────────────────────────────┐
│            Qdrant Server (localhost:6333)           │
│  - Persistent storage                               │
│  - Fast ANN search                                  │
│  - Metadata filtering                               │
└─────────────────────────────────────────────────────┘

Required Environment Variables

Setup

Option 1: Local Docker (Recommended)

Copy# Start Qdrant
docker run -p 6333:6333 -v $(pwd)/qdrant_data:/qdrant/storage qdrant/qdrant

# Verify
curl http://localhost:6333/health

Option 2: Qdrant Cloud

1. Sign up at https://cloud.qdrant.io
2. Create a cluster
3. Get API key and URL
4. Set in .env:

   CopyQDRANT_URL=https://xxx-xxx.aws.cloud.qdrant.io:6333
   QDRANT_API_KEY=your-api-key
   

Integration with TAISUN

Qdrant MCP integrates with existing memory systems:

Best Practices

1. Store with meaningful metadata

   Copy{ "topic": "react", "type": "pattern", "date": "2026-01-19" }
   

2. Use specific queries

   Copy❌ "前の話"
   ✅ "ReactのuseStateパターンについて"
   

3. Regular cleanup

   CopyOutdated knowledge should be deleted to maintain relevance
   

4. Combine with other tools

   • Context7 for docs + Qdrant for project-specific knowledge
   • GPT Researcher for external + Qdrant for internal knowledge

Sources

• Qdrant MCP Server (Official)
• Qdrant Documentation
• mcp-server-qdrant PyPI