🌐 English | Русский | 中文

MCP Server for CM3588 NAS Kit

MCP server for CM3588 NAS Kit - semantic knowledge base, change logging, live device status.

Architecture

┌────────────────────────────────────────────────────────────────┐
│  Local Machine                                                  │
│  ┌──────────────────────────────────────────────────────────┐  │
│  │  Claude Code + MCP Server                                 │  │
│  │  ├── Knowledge Base (Qdrant) - semantic search           │  │
│  │  ├── Changelog - change history                           │  │
│  │  ├── SSH client → CM3588 live status                     │  │
│  │  └── MCP Tools/Resources/Prompts                          │  │
│  └──────────────────────────────────────────────────────────┘  │
│                          │                                      │
│                          │ SSH                                  │
└──────────────────────────┼─────────────────────────────────────┘
                           ▼
┌────────────────────────────────────────────────────────────────┐
│  CM3588 NAS Kit (192.168.1.173)                                │
│  ├── RK3588 (NPU 6 TOPS)                                       │
│  ├── Wyoming/Whisper/Piper services                            │
│  └── Microphone, cameras, GPIO                                 │
└────────────────────────────────────────────────────────────────┘

Installation

1. Dependencies

# Create virtual environment
python -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -e .

2. Qdrant

docker-compose up -d

3. Configuration

cp .env.example .env

Configure in .env:

• CM3588_HOST - CM3588 IP address (192.168.1.173)
• CM3588_USER - SSH user (root)
• CM3588_SSH_KEY - path to SSH key (~/.ssh/id_ed25519)
• QDRANT_HOST, QDRANT_PORT - Qdrant (localhost:6333)

4. SSH Key

ssh-copy-id -i ~/.ssh/id_ed25519.pub root@192.168.1.173

Connect to Claude Code

Create .mcp.json in your project root:

{
  "mcpServers": {
    "cm3588": {
      "command": "/path/to/mcp-cm3588/.venv/bin/python",
      "args": ["-m", "mcp_cm3588.server"],
      "cwd": "/path/to/mcp-cm3588"
    }
  }
}

Restart Claude Code or run /mcp restart cm3588.

Tools (11 tools)

Knowledge Base

Tool	Description
save_knowledge(title, content, category, tags)	Save knowledge
search_knowledge(query, category, limit)	Semantic search
get_knowledge(entry_id)	Get entry by ID
update_knowledge(entry_id, content, append)	Update entry
list_categories()	List categories
list_knowledge(category, limit)	List entries in category

Logging

Tool	Description
log_change(what, why, details, files_changed, commands_run)	Log a change
log_solution(problem, solution, steps)	Save problem solution
get_changelog(limit)	Get change history

Documentation

Tool	Description
document_config(service_name, config_path, description)	Document config
create_howto(title, steps, category, tags)	Create step-by-step guide

Resources (14 resources)

Documentation (static)

URI	Description
docs://hardware	CM3588 specifications
docs://npu	NPU/RKNN documentation
docs://wyoming	Wyoming Protocol
docs://llm	LLM on RK3588

Live Status (from device via SSH)

URI	Description
live://system	Uptime, memory, disk, temperature
live://services	Docker containers
live://npu	NPU status, driver, load
live://network	IP addresses, ports
live://voice-pipeline	Whisper/Piper/Wake status
live://llm	LLM processes and models

Configs (from device)

URI	Description
config://whisper	Docker inspect whisper
config://piper	Docker inspect piper
config://docker-compose	docker-compose.yml files

Logs (from device)

URI	Description
logs://whisper	Whisper logs (50 lines)
logs://piper	Piper logs
logs://system	System logs

Prompts (7 templates)

Prompt	Description
setup_microphone	USB microphone setup
setup_camera	Camera setup
setup_llm_npu	Run LLM on NPU
optimize_model_npu	Optimize model for RKNN
troubleshoot_voice	Voice pipeline diagnostics
after_change	What to do after changes
document_current_state	Document current state

Knowledge Categories

• hardware - hardware, specifications
• voice-pipeline - voice pipeline (Whisper, Piper, Wake Word)
• npu - NPU, RKNN, model acceleration
• docker - Docker containers and configs
• troubleshooting - problem solutions

Usage Example

After connecting MCP server, Claude Code automatically:

1. Searches knowledge base before work:

   > How to setup microphone?
   [Claude uses search_knowledge("microphone")]
   

2. Logs changes:

   > Changed whisper config
   [Claude uses log_change()]
   

3. Checks live status:

   > What's the voice pipeline status?
   [Claude reads live://voice-pipeline]
   

4. Saves solutions:

   > Fixed NPU issue
   [Claude uses log_solution()]
   

Development

# Run server directly
python -m mcp_cm3588.server

# Linting
ruff check src/
ruff format src/

# Tests
pytest

License

MIT License - see LICENSE