Skip to content

VoxEdit MCP Server

MCP (Model Context Protocol) server enabling LLMs to control VoxEdit via its network interface.

The MCP process joins VoxEdit's network as a normal client. Its display name is taken from the MCP initialize clientInfo.name (e.g. cursor-vscode or Cursor), so you can tell IDE and agent sessions apart in the client list.

To use it, you need a running voxedit instance and start a server.

If you are modifying the settings like the port, the rcon password or anything else, you have to use the same settings for the mcp server, too. See below for args.

MCP Client Configuration

{
  "mcpServers": {
    "voxedit": {
      "command": "/path/to/vengi-voxeditmcp",
      "args": ["-set", "ve_nethostname", "127.0.0.1", "-set", "ve_netport", "10001", "-set", "ve_netrconpassword", "changeme", "-set", "ve_netpassword", "changeme"]
    }
  }
}

Screenshots

voxedit_screenshot renders a node (or the merged visible scene) to a PNG without a GL context via voxelutil::renderToImage / voxelutil::renderIsometricImage.

Large width/height values upscale the face projection and produce large MCP image payloads. Prefer omitting them (native voxel resolution) unless you need a specific size.

Useful arguments:

  • nodeUUID (optional) - model node to render; omit to merge the whole scene
  • face - front, back, left, right, up, down (default front)
  • isometric - true for an isometric view
  • width / height - optional output size in pixels
  • depthFactor - optional depth shading for orthographic renders
  • bgR / bgG / bgB / bgA - background color