coppeliasim-mcp

coppeliasim-mcp

An MCP server that lets Claude Code or any MCP client drive a running CoppeliaSim 4.10 simulation--building scenes, moving objects, creating joints and proximity sensors, running simulations, and reading sensors--without exposing arbitrary Lua execution.

Category
Visit Server

README

coppeliasim-mcp

An MCP server that lets Claude Code (or any MCP client) drive a running CoppeliaSim 4.10 simulation: build scenes, move objects, create joints and proximity sensors, run the simulation and read sensors back.

It deliberately does not expose arbitrary Lua execution. That is the main difference from other CoppeliaSim MCP servers. See Security.

Tool names are Spanish by default, with optional English and Portuguese aliases — see Tool name languages.

Not affiliated with, endorsed by, or maintained by Coppelia Robotics AG. CoppeliaSim is a trademark of Coppelia Robotics AG. This is an independent third-party integration; for the simulator itself, go to coppeliarobotics.com.


Requirements

  • CoppeliaSim 4.10 running, with the ZMQ remote API add-on active. It ships enabled by default and listens on port 23000.
  • Python 3.10 or newer.

Install

# Recommended: no clone, no virtualenv to manage
uvx coppeliasim-mcp

# Or install it
pip install coppeliasim-mcp

Register it with Claude Code:

claude mcp add coppelia -- uvx coppeliasim-mcp

Or, for any MCP client that reads a JSON config:

{
  "mcpServers": {
    "coppelia": {
      "command": "uvx",
      "args": ["coppeliasim-mcp"]
    }
  }
}

Configuration

All settings are optional environment variables. They can also live in a .env file in the working directory — see .env.example.

Variable Default What it does
COPPELIA_HOST 127.0.0.1 Host of the ZMQ remote API add-on.
COPPELIA_PUERTO 23000 Its port.
COPPELIA_DIRECTORIO_ESCENAS current working directory Only scenes under this folder can be loaded.
COPPELIA_MODO_LECTURA 0 Set to 1 to disable every tool that changes the scene.
COPPELIA_IDIOMAS es Tool-name aliases. es,en,pt or todos to add English and Portuguese.
COPPELIA_TIMEOUT 10 Seconds to wait for a reply before giving up.

Tools

Simulation controliniciar_simulacion, detener_simulacion, pausar_simulacion, estado_simulacion, tiempo_simulacion

Scenescargar_escena, cerrar_escena

Objectslistar_objetos, obtener_posicion, fijar_posicion, obtener_orientacion, fijar_orientacion, crear_primitiva, eliminar_objeto, emparentar_objeto, fijar_detectable

Jointsobtener_posicion_junta, fijar_objetivo_junta, fijar_velocidad_junta, obtener_fuerza_junta

Proximity sensorscrear_sensor_proximidad, leer_sensor_proximidad, comprobar_sensor_proximidad

Tool name languages

Tools are defined in Spanish (crear_primitiva, leer_sensor_proximidad, …). Setting COPPELIA_IDIOMAS registers aliases in English and Portuguese that point at the same functions — no duplicated logic, just more names in the catalog.

COPPELIA_IDIOMAS Tools Catalog size Cost per request
es (default) 23 15.0 KB
es,en 46 27.7 KB ~3,300 tokens
es,en,pt / todos 69 40.5 KB ~6,500 tokens

The catalog is sent to the model on every request, so aliases are off by default. Worth knowing before you turn them on: the model does not need translated names to understand you in another language. Tool names are identifiers, not user-facing text — ask for "move the cube forward" or "mova o cubo para frente" and it will reach for fijar_posicion either way. Aliases help when you want to read the catalog at a glance, or name a tool explicitly in a prompt.

Examples

examples/carrito_diferencial.py builds a complete differential drive robot with obstacle avoidance and measures whether it actually works. See examples/README.md.

Security

CoppeliaSim's Lua environment has access to os and io. A tool that runs arbitrary Lua therefore turns any prompt injection — for example text embedded in a third-party .ttt scene the model is asked to inspect — into command execution on your machine. This server has no such tool, by design.

The rest of the surface is narrow on purpose:

  • cargar_escena resolves the path (Path.resolve(strict=True)) before comparing it against COPPELIA_DIRECTORIO_ESCENAS, so ../.. and symlinks cannot escape it, and it only accepts scene extensions.
  • COPPELIA_MODO_LECTURA=1 disables every mutating tool at once.
  • The scene directory defaults to the working directory, not to your home.

Notes that save debugging time

Things about the CoppeliaSim API that are easy to get wrong, and that the tools surface directly:

  • leer_sensor_proximidad does not detect. It returns the result of the simulator's last sensor pass, so with the simulation stopped it always reports nothing. Use comprobar_sensor_proximidad to detect on demand.
  • An object must be marked detectable to be seen by a proximity sensor. That is what fijar_detectable is for, and it is the usual reason a sensor "doesn't work".
  • A wide cone pointing horizontally sees the floor before it sees your obstacle. With half-aperture a and the sensor at height h, the floor enters the cone at h / tan(a). If that is under the sensor range, the sensor reports the ground.
  • Re-parenting renumbers sibling paths. After hanging /Cylinder[1] off a chassis, /Cylinder[3] may become /Cylinder[1]. List objects again between successive emparentar_objeto calls, or resolve handles up front.
  • Parenting does not rigidly attach two dynamic shapes. Non-static shapes fall unless constrained by a joint or a force sensor.

Releasing

Publishing runs on a tag push, through .github/workflows/publicar.yml:

# bump version in pyproject.toml first, then
git tag v0.1.0 && git push --tags

The workflow refuses to publish when the tag and the version in pyproject.toml disagree, installs the built wheel on Python 3.10 and 3.13, and runs scripts/prueba_humo.py — an MCP handshake plus a check that a call with no simulator present answers instead of hanging — before it uploads anything. A PyPI version can never be overwritten or reused, so failing in CI is much cheaper than burning a version number.

It authenticates with PyPI through Trusted Publishing (OIDC), so there is no token stored in the repository secrets.

License

MIT — see LICENSE.

The MIT license covers this server only. CoppeliaSim itself is licensed separately by Coppelia Robotics AG, and this package neither includes nor redistributes any part of it — it talks to a simulator you install and license yourself.


coppeliasim-mcp (español)

Un servidor MCP para manejar una simulación de CoppeliaSim 4.10 desde Claude Code o cualquier cliente MCP: construir escenas, mover objetos, crear juntas y sensores de proximidad, correr la simulación y leer los sensores.

No expone ejecución de Lua arbitrario, a propósito. Es la diferencia principal con los otros MCP de CoppeliaSim que circulan. El Lua de CoppeliaSim tiene acceso a os e io, así que una tool de ese tipo convierte cualquier prompt injection —por ejemplo, texto dentro de una escena .ttt de terceros— en ejecución de comandos sobre tu máquina.

Sin afiliación, respaldo ni mantenimiento por parte de Coppelia Robotics AG. CoppeliaSim es una marca de Coppelia Robotics AG. Esto es una integración independiente de terceros; para el simulador, ve a coppeliarobotics.com.

Requisitos

  • CoppeliaSim 4.10 abierto, con el add-on ZMQ remote API activo. Viene habilitado por defecto, escuchando en el puerto 23000.
  • Python 3.10 o superior.

Instalación

uvx coppeliasim-mcp                              # recomendado
pip install coppeliasim-mcp                      # o instalado
claude mcp add coppelia -- uvx coppeliasim-mcp   # registrar en Claude Code

Configuración

Variables de entorno, todas opcionales. También se pueden poner en un .env en el directorio de trabajo — mira .env.example.

Variable Por defecto Para qué
COPPELIA_HOST 127.0.0.1 Host del add-on ZMQ remote API.
COPPELIA_PUERTO 23000 Su puerto.
COPPELIA_DIRECTORIO_ESCENAS directorio de trabajo Solo se pueden cargar escenas por debajo de esta carpeta.
COPPELIA_MODO_LECTURA 0 A 1 deshabilita todas las tools que modifican la escena.
COPPELIA_IDIOMAS es Alias de nombres de tools. es,en,pt o todos añade inglés y portugués.
COPPELIA_TIMEOUT 10 Segundos de espera antes de dar una respuesta por perdida.

Idiomas de los nombres de tools

Las tools se definen en español. COPPELIA_IDIOMAS registra alias en inglés y portugués sobre las mismas funciones: no duplica lógica, solo añade nombres.

COPPELIA_IDIOMAS Tools Catálogo Coste por petición
es (por defecto) 23 15.0 KB
es,en 46 27.7 KB ~3.300 tokens
es,en,pt / todos 69 40.5 KB ~6.500 tokens

El catálogo viaja en cada petición al modelo, así que los alias vienen apagados. Y conviene saber esto antes de encenderlos: el modelo no necesita los nombres traducidos para entenderte en otro idioma. Los nombres de tools son identificadores, no texto de cara al usuario — pídele "move the cube forward" o "mova o cubo para frente" y usará fijar_posicion igual. Los alias sirven para leer el catálogo de un vistazo, o para nombrar una tool explícitamente.

Ejemplos

examples/carrito_diferencial.py construye un carrito de tracción diferencial completo con evasión de obstáculos, y mide si de verdad funciona. Mira examples/README.md.

Cosas que ahorran horas de depuración

  • leer_sensor_proximidad no detecta. Devuelve el resultado del último barrido del simulador, así que con la simulación detenida siempre dice que no hay nada. Para detectar en el momento, comprobar_sensor_proximidad.
  • Un objeto tiene que estar marcado como detectable para que un sensor de proximidad lo vea. Para eso está fijar_detectable, y es la causa habitual de un sensor que "no funciona".
  • Un cono ancho en horizontal ve el suelo antes que el obstáculo. Con media apertura a y el sensor a altura h, el suelo entra en el cono a h / tan(a). Si eso queda por debajo del alcance, el sensor reporta el piso.
  • Emparentar renumera las rutas de los hermanos. Al colgar /Cylinder[1] de un chasis, /Cylinder[3] puede pasar a ser /Cylinder[1]. Vuelve a listar los objetos entre llamadas, o resuelve los handles antes de tocar la jerarquía.
  • Emparentar no une rígidamente dos cuerpos dinámicos. Las formas no estáticas se caen si no las sujeta una junta o un force sensor.

Licencia

MIT — mira LICENSE.

La licencia MIT cubre solo este servidor. CoppeliaSim se licencia por separado con Coppelia Robotics AG, y este paquete no incluye ni redistribuye ninguna parte de él: habla con un simulador que instalas y licencias tú.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured