Scumble

What Scumble MCP server does

The Scumble MCP server exposes Scumble's desktop image-editing features to MCP-compatible agents. It is designed for workflows where an agent needs to work on an image rather than only produce a standalone generation. Agents can select regions with text, create or edit image content, work with masks and layers, and export the resulting document.

Each generated result is added as its own layer. That layer can be masked, erased, blended, filtered, or adjusted before the document is flattened. Scumble also supports box-based descriptions for placing, keeping, moving, or removing objects, along with reference images identified in prompts such as @img1 and @img2.

How it works

Scumble runs as a desktop editor and provides its commands through an MCP server. An MCP client connects to the running application and can use the exposed operations in the same document and undo context as manual edits. The README describes 114 tools covering actions such as opening documents, selecting areas, generating content, changing layer matching, and exporting files.

Rendering can use a local ComfyUI instance, a remote ComfyUI deployment, Comfy Cloud, or supported provider APIs configured in Scumble. API-based rendering does not require ComfyUI. Depending on the selected provider and model, the editor can generate, inpaint, upscale, or edit image content.

Setup and configuration

Install Scumble on Windows through the Microsoft Store, winget, a GitHub installer, or a portable archive. x64 Linux builds are available as AppImage and .deb packages, although the README marks them as early CI builds that have not been run by the author. macOS support is planned but not started.

For Claude Desktop, each release includes an .mcpb package that can be installed with one click. The application also provides Help > Copy MCP registration for connecting other MCP clients. Claude Code and other MCP clients can then access the editor's commands.

Rendering requires either API credentials entered under Settings > API providers or a configured ComfyUI backend. A ComfyUI setup requires the ComfyUI-InpaintCanvas node pack and the models used by the selected recipe. Provider credentials are stored in the operating system credential store rather than a settings file.

Tools and capabilities

The Scumble MCP server supports agent-driven operations including:

• Opening and working with image documents
• Selecting image regions using text or editor interactions
• Generating new content and inpainting selected areas
• Creating masks and managing independent result layers
• Adjusting layer colour matching, blending, and filters
• Working with reference images and box-based composition instructions
• Exporting layered documents as PSD or OpenRaster

Scumble also treats filters as layers and supports text and image content within the layer stack. PSD and OpenRaster files can be reopened with their layers intact.

Limitations and notes

Scumble is free software under GPL-3.0 and does not require its own account. It is currently a young 0.1.x application, so workflows may have rough edges; the README recommends keeping backups of images. Windows is the primary supported platform. Linux packages are early builds, and macOS is not available yet.

The MCP connection controls the Scumble application; it does not replace the rendering services or models. Image generation still depends on a configured ComfyUI installation, Comfy Cloud access, or provider API keys. Provider charges may apply when using external APIs. API keys must be entered again when moving encrypted portable-install data to another Windows computer.