mcpField
C
Compositor

Compositor

Local
github.comMedia & Design8Updated 5w ago

About Compositor

Compositor MCP is an MCP server for AI agents to inspect and edit Compositor projects through the documented .comp format, without changing the app. It supports layer-level work, previews, and image exports.

Key features

  1. 01Inspect canvas and layer details
  2. 02Preview and export images
  3. 03Create projects and import layers
  4. 04Edit layers, folders, and masks
  5. 05Resize the canvas
  6. 06Preserve unsupported fields

Use cases

  1. 01Inspect a project's layer structure
  2. 02Add and position images in a project
  3. 03Set masks or layer blend modes
  4. 04Export a composite preview or layer image

MCP support

The MCP server provides clients with tools to inspect projects, edit layers, render previews, export images, and open projects in Compositor. It also exposes a resource describing the file format.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"josusanz-compositor": {
"command": "npx",
"args": [
"-y",
"compositor-mcp"
]
}
}
}
Run commandnpx -y compositor-mcp

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npmcompositor-mcp0.1.4

Tools

15

Saved tool list · updated 2w ago

ParameterTypeRequiredDescription
pathstringYesPath to the .comp project package (a directory ending in .comp)

Details

From the project README

compositor-mcp

An MCP server that lets Claude and other AI agents read and edit Compositor projects, the open-source Photoshop alternative for Mac.

It works on the documented .comp package format (a manifest.json plus one PNG per layer), so it needs no changes to Compositor itself. Edit a project with an agent, then open it in Compositor. Or the other way round: save in Compositor and let the agent inspect, retouch or render it.

What the agent can do

ToolPurpose
inspect_projectCanvas size, resolution and every layer with folder depth, opacity, blend mode, transform, mask and clipping info
list_layersCompact top-to-bottom view like the Layers panel
render_previewFlattened PNG preview returned as an image so the model can see the composition
export_flattenedFull-resolution PNG or JPEG export
export_layerCopy a layer's source pixels or mask to a file
create_projectNew document, optionally with a solid background layer
add_image_layerImport PNG, JPEG, HEIC, TIFF, WebP… as a layer, with fit/fill/stretch placement, opacity, blend mode and folder
add_folderCreate a layer folder
set_layerRename, show/hide, opacity, blend mode, position, size, rotation, flips, sampling, mask on/off
move_layerReorder layers or move them between folders
remove_layerDelete a layer or a whole folder
replace_layer_imageSwap a layer's pixels (round-trip through another tool or a generated image)
set_layer_maskAttach a grayscale mask, invert it, or clear it
resize_canvasChange the document size with an anchor, without resampling
open_in_compositorOpen the project in the Compositor app

The server also exposes a compositor://format resource describing the file format.

Related

marcushorndt/compositor-mcp takes a different route: it drives Compositor's own document model and renderer headlessly (Swift), so previews match the app exactly. This package is a plain Node server that works on the documented file format, needs no build of the app, and runs wherever Node runs. Pick whichever fits your setup; both open the same .comp files.

Install

Requires Node.js 20 or later. Compositor itself is only needed to look at the result.

Claude Code
claude mcp add compositor -- npx -y compositor-mcp
Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "compositor": {
      "command": "npx",
      "args": ["-y", "compositor-mcp"]
    }
  }
}
From source
git clone https://github.com/Josusanz/compositor-mcp
cd compositor-mcp
npm install
npm run build
claude mcp add compositor -- node /path/to/compositor-mcp/dist/src/index.js

Example session

Create a 1920×1080 project at ~/Desktop/poster.comp with a white background, add photo.jpg scaled to fill, put logo.png in the bottom right corner at 80 % opacity with Multiply, then show me a preview and open it in Compositor.

The agent calls create_project, add_image_layer twice, set_layer, render_preview and open_in_compositor. Compositor opens the document with the layers ready to keep editing.

How it maps to the Compositor format

  • Layer pixels are stored untouched as 8-bit sRGB PNGs; placement, scale, rotation and flips live in the layer transform, so edits stay non-destructive.
  • Masks are written as 8-bit grayscale PNGs without alpha, as Compositor requires.
  • Manifests are validated against the same rules as ProjectStore.swift before writing, written with sorted keys and replaced atomically.
  • Fields this tool does not understand (adjustments, effects, text, shapes, guides, anything newer) are preserved on round trip.
  • Projects are saved as format version 9 (what Compositor 1.2.6 writes); a project declaring a newer version keeps its own number and its unknown fields on round trip. Older Compositor builds reject newer versions; update the app if a file will not open.

Limitations

  • The preview renderer covers order, folders, visibility, opacity, transforms, raster masks and the blend modes sharp supports. Clipping masks, adjustment layers and layer effects are not rendered yet; the tool reports what it skipped. Compositor's own export is the reference.
  • Compositor does not watch files for changes. After the agent edits a project that is already open, close and reopen the document (or call open_in_compositor).
  • There is no live connection to the running app: selection, undo history and the viewport are session-only in Compositor and out of reach here.

Development

npm install
npm test        # builds and runs the node:test suite
npm run dev     # runs the server from TypeScript with tsx

License

MIT. Compositor is © Robbie Tilton, also MIT.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.