dreame-operator

dreame-operator — Dreame Robot Vacuum Operations

What this server does

Control a Dreame D20 Pro Plus robot vacuum through the DreameHome cloud API (and optionally local miio in hybrid mode). Provides device status telemetry, LIDAR map retrieval with room/path/zone data, cleaning controls, and an agentic workflow that plans multi-step cleaning goals via LLM sampling.

Tools

dreame_tool(operation=...) — portmanteau

• status — full telemetry: battery, state, cleaned area, cleaning time, charging, fan speed
• battery — battery percentage only
• map — LIDAR map summary: rooms, robot position, movement trail, restricted zones
• start_clean, stop, pause, go_home, find_robot — navigation & control

dreame_help(category=...)

Multi-level help. Categories: status, map, control, connection, agentic.

dreame_agentic_workflow(goal=...)

LLM plans and executes multi-step goals via MCP sampling (e.g. "check battery and start cleaning if above 20%"). The planner calls back into dreame_tool.

dreame_shutdown()

Gracefully disconnects the client and stops the server.

Best practices

1. Start with dreame_tool(operation='status') to verify connectivity before control ops.
2. Use dreame_agentic_workflow for compound goals; use dreame_tool directly for single commands.
3. map may take up to 60s on slow cloud links — the server serializes map fetches.
4. Control commands execute through the cloud when no local IP is set; hybrid mode (DREAME_IP + DREAME_USER/PASSWORD) gives fast local control plus cloud maps.

Configuration

• DREAME_USER / DREAME_PASSWORD / DREAME_COUNTRY — DreameHome cloud credentials (required for maps)
• DREAME_DID — optional device id (auto-discovered when a single device)
• DREAME_IP / DREAME_TOKEN — optional local miio (null-token trick when token blank)
• DREAME_MCP_PORT — backend port (default 10894)
• DREAME_REF_PATH — path to the Tasshack dreame-vacuum reference clone