Gazebo MCP Server
Gazebo and gz-sim robotics orchestration for AI agents with offline mocking capabilities.
Data last scanned today · Reviewed today
Overview
Gazebo MCP Server allows AI agents to interface directly with Gazebo or gz-sim environments. It provides capabilities to manage simulation worlds, models, and poses, along with the ability to pause or step through simulations programmatically. A key feature of this server is its offline mock mode. This enables users to perform CI testing and demonstrations without requiring a full local installation of the Gazebo simulator, facilitating development in restricted or lightweight environments.
Our verdict
With a security score of 43/100, this project is classified as high-risk. A primary security concern is the lack of a read-only mode, as every installation grants full write access to the simulation environment. Because this is a community-maintained project and not an official vendor tool, users should exercise caution and review the codebase before integration, particularly in environments where unauthorized simulation state changes could be disruptive.
- Zero direct dependencies minimize the software supply chain surface
- Includes a dedicated offline mock mode for CI testing
- Fully documented installation for multiple AI host environments
- MIT licensed open source repository
- Community-maintained status lacks official vendor oversight
- Full write access is granted without a read-only configuration
- Authentication mechanisms remain unreviewed for security
Tools
| Tool | Description | Risk |
|---|---|---|
| gazebo_doctor | Connectivity / sim health | low |
| gazebo_seed_demo | Reset mock shapes world | medium |
| gazebo_world_info | World name, paused, sim time | low |
| gazebo_world_list | Available mock/live worlds and active world | low |
| gazebo_list_models | Models in the world | low |
| gazebo_spawn | Model lifecycle | medium |
| gazebo_delete | Model lifecycle | high |
| gazebo_set_pose | Pose control | medium |
| gazebo_get_pose | Pose control | low |
| gazebo_pause | Clock control | medium |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf | ||||
| ChatGPT |
Frequently asked questions
›Do I need to have Gazebo installed to use this MCP server?
No, the server includes an offline mock mode that simulates worlds and models without requiring a Gazebo installation.
›What is the difference between mock mode and live mode?
Mock mode (default) uses an in-memory world graph for offline operations, while live mode forwards commands to a local Gazebo bridge.
›Which Python version is required?
The server requires Python 3.11 or higher.
›How do I configure the server for use with MCP-compatible agents?
You can configure most agents by pointing them to the stdio command 'gazebo-mcp' with the argument 'serve'.
›Can I restrict which models can be spawned in live mode?
Yes, you can use the GAZEBO_MCP_SPAWN_ALLOWLIST environment variable to specify allowed model types.
›How can I verify the server health or troubleshoot connectivity?
You can use the 'gazebo-mcp doctor' CLI command to check connectivity and simulation health.
Badge
Maintain this server? Add the live badge to your README.