FeyaGate Skill — MCP Smart Home Gateway for AI Agents

MCP-based multi-platform smart home gateway supporting Xiaomi, Tuya, Midea, eWeLink, Serial, and GPIO. Compatible with OpenClaw, Claude Code, Hermes, Codex, Windsurf, Copilot, 小智AI and other MCP-compatible AI agents.

Before Using Any Tool

Run this check before calling any FeyaGate MCP tool:

feyagate status

• If status is RUNNING → proceed with tool calls.
• If status is STOPPED or command not found → run feyagate start first. If feyagate is not installed, run pip install feyagate-skill && feyagate setup && feyagate start.

Dual API endpoints:

Note: The PC proxy (miloco-mcp-server) provides all gateway MCP tools plus additional extensions (camera P2P, Xiaomi auth, etc.).

Quick Start

pip install feyagate-skill       # Step 1: Install Python package
feyagate setup                   # Step 2: Download MCP server binary (~30MB)
feyagate start                   #        Start service (localhost:38080)
feyagate install-claude          # Step 3: Register with your AI agent
feyagate auth                    # Step 4: Authorize smart home platform

Other agent install commands: install-cursor, install-openclaw, install-hermes, install-codex, install-windsurf, install-copilot

Full step-by-step guide: QUICKSTART.md

Server Lifecycle (CLI)

Manual Scripts (Advanced)

Cross-Platform Tools

Unified interfaces that work across all platforms. System auto-detects device platform.

Parameter convention: device/specs uses deviceId (camelCase); platform-specific tools (xiaomi/*, tuya/*, etc.) use device_id (snake_case).

Sub-Skill Loading Guide (for AI agents)

This skill uses a modular structure. The main skill defines cross-platform tools; platform-specific tools are in sub-skill files.

Loading strategy:

1. Always load this main SKILL.md first — it provides the core cross-platform tools and MCP endpoint info.
2. Load sub-skills on demand based on the user's intent. Each sub-skill's trigger keywords (listed in frontmatter sub_skills) indicate when to load it.
3. Cross-platform workflow (no sub-skill needed): device/list → device/specs → identify platform → load corresponding sub-skill for control operations.

License System

• Free edition: Xiaomi platform only (device control, cameras, XiaoAI, MCP proxy)
• Licensed edition: All platforms (Xiaomi + Tuya + Midea + eWeLink)
• tuya/set_property, midea/set_property, ewelink/set_property return license_required on free edition
• get_properties and all other tools work without license

Configuration

config/config.yaml:

server:
  http_port: 38080
  ws_port: 8765
  bind_address: "0.0.0.0"
auth:
  cloud_server: "cn"      # cn / de / sg / us / ru / i2
  token_file: "data/auth_token.json"
camera:
  frame_interval: 500
  buffer_max_size: 20
  buffer_ttl: 300
  jpeg_quality: 90
tuya:
  token_file: "data/tuya_token.json"
midea:
  token_file: "data/midea_token.json"
ewelink:
  token_file: "data/ewelink_token.json"
xiaozhi:
  endpoint: ""            # ws:// or wss:// (empty = disabled)

Troubleshooting

Full API reference: FeyaGate_MCP_API.md, FeyaGate_HTTP_API.md