Robotics MCP Server
Servidor MCP de Robótica
Autor: FlowEngineer sandraschi
Un servidor MCP para robots físicos (aspiradoras Dreame, Yahboom ROS, Elegoo, ), Gazebo y pilas virtuales (Unity, VRChat). FastMCP, stdio + HTTP. Reinicios opcionales con watchfiles para desarrollo o ejecuciones desatendidas (no es un SLA).
Doc | Contenido |
Requisitos previos, hardware, instalación, empaquetado, Cursor, configuración, MCPs compuestos, pruebas, solución de problemas | |
Ejemplos de herramientas, interfaz web, endpoints HTTP | |
Estándares de seguridad, legales y operativos para la flota de androides | |
Unity, VRChat, servidores MCP pares cuando necesites la pila completa | |
Reinicio automático y comprobaciones de estado |
Estado: Industrial v1.4.1 (14-04-2026) | Cumplimiento SOTA v1.4.1 | Protocolo Benny asegurado.
[
](https://goFastMCP 3.1.0com)
Instalación
uvx robotics-mcppip install robotics-mcpClonación para desarrollo: ver Instalación en el README técnico.
Related MCP server: ROS MCP
Contribución
Ver CONTRIBUTING.md.
🛡️ Pila de Calidad Industrial
Este proyecto cumple con los estándares industriales SOTA 14.1 para la orquestación de agentes de alta fidelidad:
Python (Núcleo): Ruff para linting y formateo. Tolerancia cero para sentencias
printen los manejadores principales (T201).Webapp (UI): Biome para linting de sub-milisegundos. Aplicación estricta de
noConsoleLog.Cumplimiento del protocolo: Aislamiento endurecido de
stdout/stderrpara garantizar una comunicación JSON-RPC resistente a fallos.Automatización: Recetas de Justfile para todas las operaciones de flota (
just lint,just fix,just dev).Seguridad: Auditorías automatizadas mediante
banditysafety.
Licencia
MIT ver LICENSE.
Agradecimientos
FastMCP; comunidad ROS; Unity3D, VRChat, World Labs Marble/Chisel; colaboradores del ecosistema MCP.
Available Tools
25 toolsdreame_controlA
Dreame D20 Pro vacuum control with conversational responses.
Provides unified interface for Dreame vacuum operations: status, battery, water level, suction, LIDAR map, cleaning programs, zone/spot/room cleaning, and configuration.
Operations:
get_status: Battery, water, suction, charging state, device status
start_cleaning / stop_cleaning: Full-house or spot cleaning
return_to_dock: Return to charging dock
move: Manual drive (rotation, velocity)
play_sound: Locate robot
set_suction_level: Suction power 1-4
set_water_volume: Water flow 1-3
set_mop_humidity: Mop pad humidity 1-3
get_map: LIDAR map data (rooms, walls, robot/charger positions)
export_map: Export LIDAR map to OBJ/PLY/Unity/Blender formats
clean_zone / clean_spot / clean_room: Targeted cleaning
get_cleaning_history / clear_error: History and error recovery
Inputs: operation (required), token, zones, spot_x, +11 more. Returns: object
dreame_onboardA
Dreame D20 Pro Plus — DreameHome cloud setup (no local token).
This server uses the DreameHome cloud API. Set environment variables and optionally the Tasshack ref clone path. No network discovery or miio token needed.
Returns: Setup instructions and env var reference.
Inputs: no named parameters. Returns: object
drone_controlC
Core drone flight control operations with conversational responses.
Provides unified control interface for PX4/ArduPilot drones via MAVLink protocol or direct firmware integration. Supports both autonomous and manual flight modes with rich conversational responses for natural AI interaction.
Inputs: drone_id (required), operation (required), mode, altitude, +5 more. Returns: object
drone_flight_controlC
Manage advanced drone flight control, missions, and parameters with conversational responses.
Provides fine-grained control over PX4/ArduPilot flight modes, autonomous mission execution, and drone parameter tuning for advanced flight operations with intelligent mission planning and parameter optimization recommendations.
Inputs: drone_id (required), operation (required), mode, mission_id, +3 more. Returns: object
drone_streamingC
Manage drone video streaming and recording with conversational responses.
Controls FPV video feeds, RTSP/WebRTC streaming, and onboard recording. Supports multiple protocols for real-time monitoring and data capture with intelligent quality adaptation and conversational status updates.
Inputs: drone_id (required), operation (required), bitrate, quality, +2 more. Returns: object
environmental_logistics_handlerC
Handle specialized environmental and substrate logistics.
Covers everything from textiles and darkroom film to wildlife search & rescue, autonomous security patrols, pet logistics, pool supervision, human health, and mural painting.
Inputs: operation (required), material_type (required), action, force_n, +7 more. Returns: object
gazebo_modelsB
Manage Gazebo Fuel simulation models - browse, download, and spawn.
Browse, download, and spawn 3000+ free models from Gazebo Fuel (fuel.gazebosim.org) into your Gazebo simulation environment.
Inputs: operation (required), x, y, z, +7 more. Returns: object
noetix_infoA
Noetix Bumi humanoid robot — features, specs, and SDK links.
Operations: info — return features, specs, GitHub and docs links (default)
Returns: success, message, and data (specs, features, github, docs).
Inputs: operation. Returns: object
robot_behaviorA
Robot behavior control portmanteau - Animation, camera, navigation, and manipulation.
PORTMANTEAU PATTERN: Consolidates animation, camera, navigation, and manipulation operations into a single unified tool. This reduces tool explosion while maintaining full functionality across all behavior categories.
CATEGORIES:
animation: Animation and pose control
camera: Camera feed and visual control
navigation: Path planning and navigation
manipulation: Arm and gripper control
ANIMATION ACTIONS:
animate_wheels: Rotate wheels during movement (Scout mecanum wheels)
animate_movement: Play movement animations (walk, turn, etc.)
set_pose: Set robot pose (sitting, standing, etc. for Unitree)
play_animation: Play custom animations
stop_animation: Stop current animation
get_animation_state: Get current animation state
CAMERA ACTIONS:
get_camera_feed: Get live camera feed (physical Scout camera)
get_virtual_camera: Get Unity camera view from robot perspective
set_camera_angle: Adjust camera angle
capture_image: Capture still image
start_streaming: Start video stream
stop_streaming: Stop video stream
get_camera_status: Get camera status and settings
NAVIGATION ACTIONS:
plan_path: Plan path from A to B (A* or RRT)
follow_path: Execute planned path
set_waypoint: Set navigation waypoint
clear_waypoints: Clear waypoint list
get_path_status: Check path execution status
avoid_obstacle: Dynamic obstacle avoidance
get_current_path: Get current path being followed
MANIPULATION ACTIONS:
move_arm: Move arm to target joint positions or end-effector pose
set_joint_positions: Set individual joint positions (dict of joint_name: angle)
set_end_effector_pose: Move end-effector to target pose (position + orientation)
get_arm_state: Get current arm joint positions and end-effector pose
open_gripper: Open gripper fully
close_gripper: Close gripper fully
set_gripper_position: Set gripper position (0.0 = open, 1.0 = closed)
get_gripper_state: Get current gripper position and force feedback
move_to_pose: Move arm to target pose with IK (inverse kinematics)
home_arm: Return arm to home/rest position
Inputs: action (required), category (required), robot_id (required), loop, +19 more. Returns: object
robot_controlC
Unified robot control with conversational responses.
Provides a single interface for controlling physical robots, virtual robots, and specialized devices with intelligent routing and rich conversational responses. Primary: Dreame D20 Pro vacuum. Also supports Yahboom ROSMASTER, Moorebot Scout, Unitree (when hardware available), and virtual robots in Unity/VRChat.
PORTMANTEAU PATTERN RATIONALE: Instead of creating separate tools for each robot type and operation, this tool consolidates all robot control operations into a single interface. This design:
Prevents tool explosion (15+ tools -> 1 tool) while maintaining full functionality
Enables seamless switching between physical and virtual robots
Provides consistent error handling and safety protocols across all robot types
Supports conversational AI interaction with context-aware responses
Follows FastMCP 2.13+ best practices for feature-rich MCP servers
SUPPORTED ROBOT TYPES:
Dreame D20 Pro: Primary platform - LIDAR vacuum with mapping, zone cleaning, mop, auto-empty
Yahboom Robots: ROSMASTER series with AI, navigation, and optional robotic arms
Moorebot Scout: ROS1 wheeled robot
Elegoo Robots: ROS-on-PC robots with serial communication
Unitree Go2/G1: Quadrupedal robots (when hardware available)
Virtual Robots: Unity3D/VRChat robots
Hue HomeAware: Philips Hue Bridge Pro with RF-based movement detection
SUPPORTED OPERATIONS:
Universal: "get_status", "move", "stop"
Physical Robots: "return_to_dock", "stand", "sit", "walk", "sync_vbot"
Yahboom: "home_patrol", "camera_capture", "arm_move", "gripper_control", "navigate_to", "ai_query"
Elegoo: "emergency_stop" (ROS-on-PC serial communication)
Dreame: "start_auto_empty", "stop_auto_empty", "start_self_clean", "stop_self_clean", "set_suction_level", "set_water_volume", "set_mop_humidity", "clean_zone", "clean_spot", "start_mapping", "rename_room", "set_cleaning_sequence", "set_restricted_zones", "get_cleaning_history", "clear_error"
Hue HomeAware: "hue_get_movement_events", "hue_get_sensor_status", "hue_get_movement_zones"
Inputs: action (required), robot_id (required), x, y, +19 more. Returns: object
robotics_agentic_workflowC
Execute a complex, multi-step robotics workflow leveraging MCP sampling.
Inputs: objective (required), thought_process (required). Returns: result
robotics_fab_artA
Fab art fleet bridge via inkscape-mcp HTTP (DXF, laser dots, Gazebo schematics).
Operations: inkscape_status: Probe inkscape-mcp and list fab presets. batch_dxf_export: Run inkscape_fab_art batch_dxf_export on input_dir. batch_laser_dots: Generate laser dot SVG batch from preset. gazebo_schematic: SVG to PNG schematic for model documentation. stage_fab_paths: Stage DXF/SVG/PNG outputs for robotics workflows. push_gimp_schematic: Export schematic and push through gimp validation.
Inputs: operation (required), gimp_url, svg_path, input_dir, +5 more. Returns: object
robotics_sim_artB
Sim-art fleet bridge via gimp-mcp HTTP (Gazebo icons, VRChat, avatar thumbs).
Uses HTTP tool calls instead of mounting gimp-mcp stdio (avoids protocol hangs).
Operations: gimp_status: Probe gimp-mcp health and list sim-art templates. batch_gazebo_icons: Run gimp_sim_art_tool gazebo_model_icons on input_dir. import_gazebo_models: Batch import thumbnails into models_root (auto_import). batch_vrchat_icons: Run vrchat_icon_batch on input_dir. avatar_thumbnail: Push icon_path to avatar via avatar_manager set_thumbnail.
Inputs: operation (required), gimp_url, avatar_id, icon_path, +6 more. Returns: object
robotics_systemB
System management portmanteau for Robotics MCP.
PORTMANTEAU PATTERN RATIONALE: Instead of creating 3 separate tools (help, status, list_robots), this tool consolidates related system operations into a single interface. This design:
Prevents tool explosion (3 tools -> 1 tool) while maintaining full functionality
Improves discoverability by grouping related operations together
Reduces cognitive load when working with system management tasks
Enables consistent system interface across all operations
Follows FastMCP 2.13+ best practices for feature-rich MCP servers
SUPPORTED OPERATIONS:
help: Get comprehensive help information about the server and its tools
status: Get server status with connectivity tests and robot counts
list_robots: List all registered robots with optional filtering
Inputs: operation (required), is_virtual, robot_type. Returns: object
robot_manufacturingB
Manufacturing equipment control with conversational responses.
Provides unified control over 3D printers, CNC machines, and laser cutters with rich conversational responses for natural AI interaction. Supports complete manufacturing workflows from file upload to finished product.
PORTMANTEAU PATTERN RATIONALE: Instead of creating separate tools for each device type and operation, this tool consolidates manufacturing operations into a single interface. This design:
Prevents tool explosion (9+ tools -> 1 tool) while maintaining full functionality
Enables consistent manufacturing workflow across different device types
Provides unified error handling and safety protocols
Supports conversational AI interaction with rich response formats
Follows FastMCP 2.13+ best practices for feature-rich MCP servers
SUPPORTED OPERATIONS:
3D Printers: Print control, monitoring, maintenance (OctoPrint/Moonraker)
CNC Machines: Milling, drilling, cutting operations
Laser Cutters: Vector cutting, engraving, material processing
Inputs: action (required), category (required), device_id (required), device_type (required), +7 more. Returns: object
robot_modelB
Robot model management portmanteau for Robotics MCP.
PORTMANTEAU PATTERN RATIONALE: Instead of creating 4 separate tools (create, import, export, convert), this tool consolidates related model operations into a single interface. This design:
Prevents tool explosion (4 tools -> 1 tool) while maintaining full functionality
Improves discoverability by grouping related operations together
Reduces cognitive load when working with robot models
Enables consistent model interface across all operations
Follows FastMCP 2.13+ best practices for feature-rich MCP servers
SUPPORTED OPERATIONS:
create: Create robot 3D model from scratch using Blender MCP
import: Import robot 3D model into Unity/VRChat/Resonite project
export: Export robot model from Unity to file format
convert: Convert robot model between formats
spz_check: Check .spz conversion tool availability
spz_convert: Convert .spz file to .ply or other format
spz_extract: Extract metadata from .spz file
spz_install: Install Unity Gaussian Splatting plugin (alternative to .spz)
execute_script: Execute custom Blender Python script (like Gemini-generated scripts)
list_scripts: List all available scripts in the repository by category
execute_repository_script: Execute a script from the repository with optional parameters
Inputs: operation (required), format, script, category, +19 more. Returns: object
robot_virtualB
Virtual robot lifecycle and operations portmanteau.
PORTMANTEAU PATTERN: Consolidates virtual robot CRUD operations and virtual robotics operations into a single unified tool. This reduces tool explosion while maintaining full functionality for virtual robot management.
CRUD OPERATIONS:
create: Create/spawn and register a new virtual robot
read: Get details of an existing virtual robot
update: Modify virtual robot properties (scale, position, metadata, etc.)
delete: Remove and unregister a virtual robot
list: List all virtual robots with optional filtering
VIRTUAL ROBOT OPERATIONS:
spawn: Spawn robot in Unity/VRChat scene (alias for create)
load_environment: Load Marble/Chisel environment into scene
get_status: Get virtual robot status
get_lidar: Get virtual LiDAR scan (Unity physics raycast)
set_scale: Scale robot size (for size testing)
test_navigation: Test pathfinding in environment
sync_with_physical: Sync vbot state with physical bot
Inputs: operation (required), scale, metadata, platform, +8 more. Returns: object
sim_fleet_backendsA
List all registered sim backends and their capabilities (offline registry).
Inputs: no named parameters. Returns: object
sim_fleet_routeA
Route a simulation task to the best available backend.
Probes all running sim MCPs, picks the optimal one based on task keywords (sensors → Gazebo, rendering → Isaac, locomotion → MuJoCo, TRON 1/Oli → LimX), and returns the backend URL and metadata. The caller then makes the actual API call to that backend.
Inputs: task (required), preferred_backend. Returns: object
sim_fleet_statusA
Probe all registered simulation MCP backends and report availability.
Checks: mujoco-mcp (MuJoCo), gazebo-mcp (Gazebo), isaac-mcp (Isaac Sim), limx-robotics-mcp (LimX). Returns health data, strengths, and ports for each.
Inputs: no named parameters. Returns: object
sim_marketplace_info_toolA
Get full metadata for a specific robot model from the marketplace.
Returns file paths (MJCF, URDF), compatible sim backends, joint/actuator counts, and source URLs.
Return Format
{"success": bool, "message": str, "model": {"id": str, "name": str, ...} | None}
Examples
sim_marketplace_info_tool(model_id="unitree_go2") sim_marketplace_info_tool(model_id="limx_oli_d04")
Inputs: model_id (required). Returns: object
sim_marketplace_search_toolA
Search the robot model marketplace for available models.
Filters by keyword, tags, sim backend, or robot type. When exact keyword match returns nothing, falls back to a local LLM (Ollama) to suggest the closest catalog entry.
Return Format
{"success": bool, "message": str, "matches": [{ "id", "name", "type", "tags", "sim_backends", "formats", "joint_count", "actuator_count" }], "total": int, "llm_fallback_used": bool}
Examples
sim_marketplace_search_tool(query="go2") sim_marketplace_search_tool(tags=["humanoid", "unitree"]) sim_marketplace_search_tool(sim_backend="gazebo")
Inputs: tags, type, query, sim_backend. Returns: object
vbot_crudA
CRUD operations for virtual robots (vbots).
This tool provides complete lifecycle management for virtual robots:
Create: Spawn and register a new virtual robot
Read: Get details of an existing virtual robot
Update: Modify virtual robot properties (scale, position, metadata, etc.)
Delete: Remove and unregister a virtual robot
List: List all virtual robots with optional filtering
Supported robot types:
"scout": Moorebot Scout (mecanum wheels, indoor)
"scout_e": Moorebot Scout E (tracked, waterproof, outdoor)
"go2": Unitree Go2 (quadruped)
"g1": Unitree G1 (humanoid with arms)
"yahboom": Yahboom ROSMASTER series (AI-enabled wheeled robots)
"drone": Generic PX4/ArduPilot drones
"px4_quad": PX4 quadcopter drones
"ardupilot": ArduPilot-based drones
"robbie": Robbie from Forbidden Planet (classic sci-fi robot)
"custom": Custom robot type (requires model_path)
Inputs: operation (required), scale, metadata, platform, +4 more. Returns: object
workflow_managementC
Comprehensive workflow management operations.
OPERATIONS:
create: Create new workflow (requires workflow_data)
read: Get workflow details (requires workflow_id)
update: Update workflow (requires workflow_id, workflow_data)
delete: Delete workflow (requires workflow_id)
list: List all workflows (filterable by category, tags, search)
execute: Execute workflow (requires workflow_id, variables)
status: Get execution status (requires execution_id)
templates: List available workflow templates
import: Import workflow from JSON (requires workflow_data)
export: Export workflow to JSON (requires workflow_id)
Inputs: operation (required), tags, search, category, +5 more. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 25 tools
Tool boundaries are severely unclear: `robot_virtual` and `vbot_crud` both provide vbot CRUD, `robot_control` and `dreame_control` both expose Dreame vacuum commands, and `drone_control`, `drone_flight_control`, and `drone_navigation` overlap heavily on flight and mission management. The portmanteau-style operation lists make it hard for an agent to know which tool is canonical for a given action.
Most names are snake_case, but they do not follow a consistent action-oriented pattern: some use `_control`, some `_management`, some `_info`, some `_tool`, and some are arbitrary nouns like `robot_virtual`, `vbot_crud`, or `noetix_info`. This makes it difficult to predict tool names or to infer which tool will handle a specific operation.
The 25-tool count is at the heavy end of a reasonable range for such a broad robotics server, but several tools are near-duplicates or one-off info helpers. The set would be easier to navigate if the overlap-free core were consolidated.
The surface covers many important areas: virtual robot lifecycle, model management, workflow CRUD, Dreame vacuum controls, drone subsystems, and marketplace lookup. However, `robot_manufacturing` lacks concrete operations, sim backend lifecycle management is only partially exposed, and `robotics_agentic_workflow`/`environmental_logistics_handler` are opaque additions that create coverage gaps.
Maintenance
Related MCP Connectors
Robot embodied-AI cloud for GPU training, inference, benchmarks, and robot data collection.
Zero-auth ROS 2 Humanoid, Q3 Proxemics, 3-Level Voice Arbiter, A2A-MCP Bridge & W3C VC 2.0.
Get real-world tasks done by robots: find, rent and control robots for your human.
Generate 3D models, worlds, PBR materials, and coordinated asset packs from AI agents.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables control of ROS/ROS2 robots through natural language commands by translating LLM instructions into ROS topics and services. Supports cross-platform WebSocket-based communication with existing robot systems without requiring code modifications.MIT
- AlicenseAqualityDmaintenanceEnables controlling robots in ROS environments through natural language, supporting topics, services, actions, and GUI tools.2436MIT
- AlicenseNot gradedqualityBmaintenanceProvides a runtime for autonomous virtual robot brains, sitting between game engines and LLM inference to offer perception ingestion, LLM-driven reasoning, episodic memory, personality engine, motor output, and multi-bot registry.2MIT
- AlicenseNot gradedqualityDmaintenanceEnables large language models to interact with ROS robots seamlessly, allowing natural language control, real-time sensor monitoring, and autonomous task execution without modifying existing robot code.Apache 2.0