Qdrant Memory MCP Server V2

MCP Server for semantic memory storage with Qdrant vector database and two-stage retrieval.

Features

• ✅ Two-stage retrieval (preview → full content)
• ✅ Role-based memory collections (universal, backend, frontend, etc.)
• ✅ Semantic search via OpenAI embeddings
• ✅ Remote or local Qdrant support
• ✅ Project-specific and global memory storage
• ✅ CRUD operations for memories

Quick Start

One-Command Installation (Recommended)

Install everything (MCP server, Docker + Qdrant, skills, subagent, config) with one command:

git clone https://github.com/hungson175/deploy-memory-tools.git
cd deploy-memory-tools
./install.sh

What it installs:

• ✅ Python package (qdrant-memory-mcp)
• ✅ Docker container running Qdrant
• ✅ Skills (coder-memory-recall, coder-memory-store)
• ✅ Memory-only subagent
• ✅ MCP server configuration in ~/.claude.json
• ✅ Environment setup (.env file)

Requirements:

• Docker (running)
• Python 3.10+
• Claude Code

After installation:

1. Add your OpenAI API key to .env
2. Restart Claude Code
3. Memory system auto-triggers on complex tasks

Manual Installation (Advanced)

<details> <summary>Click to expand manual installation steps</summary>

1. Install Python Package

# Using uv (recommended)
uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
uv pip install -e .

# Or using pip
pip install -e .

2. Start Qdrant

docker-compose up -d

3. Configuration

Copy .env.example to .env and configure:

cp .env.example .env
# Edit .env with your OpenAI API key

4. Install Skills & Subagent

# Copy skills
cp -r skills/coder-memory-recall ~/.claude/skills/
cp -r skills/coder-memory-store ~/.claude/skills/

# Copy subagent
cp -r subagents/memory-only ~/.claude/subagents/

5. Update ~/.claude.json

See "Configuration for Claude Code" section below.

</details>

Uninstall

./uninstall.sh

Project Structure

deploy-memory-tools/
├── src/qdrant_memory_mcp/     # Main package
│   ├── __main__.py            # Entry point + server logic
│   ├── config.py              # Configuration
│   └── utils/                 # Helper utilities
├── skills/                    # Claude Code skills
├── subagents/                 # Memory-only subagent
├── docs/                      # Documentation & backlog
├── install.sh                 # One-command installation
├── uninstall.sh               # Uninstallation
└── pyproject.toml             # Python project config

MCP Tools

The server provides 7 MCP tools:

1. list_collections - List all memory collections
2. search_memory - Semantic search (returns previews)
3. get_memory - Get full memory by ID
4. batch_get_memories - Get multiple memories efficiently
5. store_memory - Store new memory
6. update_memory - Update existing memory
7. delete_memory - Delete memory

Memory Collections

Global Collections (Role-Based)

• universal-patterns - Cross-domain patterns
• backend-patterns - Backend engineering
• frontend-patterns - Frontend development
• quant-patterns - Quantitative finance
• devops-patterns - DevOps & infrastructure
• ml-patterns - Machine learning
• security-patterns - Security engineering
• mobile-patterns - Mobile development

Project Collections

• proj-{project-name} - Project-specific memories

Development

Running Tests

pytest tests/

Code Formatting

black src/ tests/
ruff check src/ tests/

Type Checking

mypy src/

Configuration for Claude Code

Add to your ~/.claude.json:

{
  "mcpServers": {
    "memory": {
      "type": "stdio",
      "command": "python3",
      "args": [
        "/path/to/deploy-memory-tools/src/qdrant_memory_mcp/__main__.py"
      ],
      "env": {
        "QDRANT_URL": "http://your-server:6309",
        "OPENAI_API_KEY": "sk-proj-..."
      }
    }
  }
}

Documentation

• Project Documentation

License

MIT