Congatudo MCP Server

HTTP MCP server (SSE transport) that exposes your Congatudo-powered Cecotec Conga robot vacuum as 26 tools for AI agents (Claude, Cursor, OpenAI Agents SDK, etc.).

Quick Start

# 1. Clone and configure
cp .env.example .env
# Edit .env → set CONGATUDO_HOST to your robot's IP

# 2. Build and start
docker compose up -d

# 3. Connect any MCP client to:
#    http://<host>:8114/sse

Related MCP server: Roborock MCP Server

Environment Variables

Tools

State

• get_robot_state — full state (status, battery, attachments, map)

• get_robot_state_attributes — status / battery / attachment attributes

• get_robot_map — raw map JSON

Control

• basic_control(action) — start / pause / stop / home

• locate_robot() — plays a sound to find the robot

• manual_control(action, movement_command?) — check / enable / disable / move

Cleaning Modes

• clean_zone(zones, iterations?) — clean rectangular zone(s)

• clean_segments(segment_ids, iterations?, custom_order?) — clean specific rooms

• go_to_location(x, y) — send robot to map coordinates

Settings (fan + water)

• get_fan_presets() / set_fan_speed(name) — suction power

• get_water_presets() / set_water_usage(name) — mopping water flow

Consumables

• get_consumables() — wear levels

• reset_consumable(type, sub_type?) — reset after replacement

Timers (cron-based)

• get_timers() / create_timer(...) / update_timer(id, ...) / delete_timer(id) / toggle_timer(id)

Do Not Disturb

• get_dnd_config() / set_dnd_config(enabled, start, end) — quiet hours

System

• get_capabilities() — list what your robot supports

• get_robot_info() — manufacturer, model, implementation

• get_wifi_status() — SSID, RSSI, frequency, IPs

• get_system_info() — hostname, arch, uptime, CPU, memory

Call get_capabilities() first — if a capability isn't listed, its tool will return a clean error.

Testing

A bash script validates the MCP connection and exercises 5 key tools:

./test_get_capabilities.sh
# or against a remote host:
./test_get_capabilities.sh http://192.168.1.100:8114

It performs the full MCP handshake, calls get_capabilities, get_fan_presets, get_consumables, locate_robot (beeps the robot), and basic_control(action=home) (sends robot to dock). Output:

--- MCP Congatudo Test ---
1. get_capabilities              PASS  (21 capabilities)
2. get_fan_presets               PASS  (4 presets)
3. get_consumables               PASS  (4 consumables)
4. locate_robot                  PASS  (robot beeps)
5. basic_control (HOME)          PASS  (robot returns to dock)
--- Results: 5 passed, 0 failed ---

MCP Integration

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
	  "Congatudo": {
      "command": "mcp-proxy.exe",
      "env": {
        "SSE_URL": "http://localhost:8114/sse"
      }		
    }
  }
}

Cursor

In Cursor Settings → MCP Servers → Add new:

• Name: congatudo

• Type: SSE

• URL: http://192.168.1.100:8114/sse

Opencode

Add to your opencode.json or .opencode.json:

{
  "mcp": {
    "congatudo": {
      "transport": "sse",
      "url": "http://192.168.1.100:8114/sse"
    }
  }
}

OpenAI Agents SDK (Python)

from agents import Agent, Runner
from agents.mcp import MCPServerSse

async with MCPServerSse(
    name="Congatudo",
    params={"url": "http://192.168.1.100:8114/sse"},
) as server:
    agent = Agent(name="Vacuum", mcp_servers=[server])
    await Runner.run(agent, "Start cleaning the kitchen")

Any MCP Client

Point your client to:

http://<host>:8114/sse

Then call get_capabilities() first to discover what your robot supports.

Docker Commands

docker compose build      # Build the image
docker compose up -d      # Start in background
docker compose logs -f    # Follow logs
docker compose down       # Stop