Overview
What is Photoshop MCP Server?
A Model Context Protocol (MCP) server that enables remote control of Adobe Photoshop on macOS and Windows using FastAPI and multiple backend bridges (AppleScript, PowerShell, UXP).
How to use Photoshop MCP Server?
Install via pip (pip install photoshop-mcp-server) and start the server with photoshop-mcp-server start. Specify a backend bridge mode with --bridge-mode (e.g., applescript, powershell, uxp). Use REST or WebSocket endpoints to control Photoshop.
Key features of Photoshop MCP Server
- Cross‑platform support (macOS and Windows)
- Multiple backend bridges (AppleScript, PowerShell, UXP plug‑in)
- Thumbnail generation from PSD files (REST and WebSocket streaming)
- Cluster mode for managing multiple Photoshop instances
- LLM‑powered auto retouch using LiteLLM
- Performance optimizations (script caching, parallel processing)
Use cases of Photoshop MCP Server
- Batch image processing and automated editing workflows
- Integrating AI assistants with Photoshop for retouching tasks
- Running repetitive actions across many PSD files via Cluster mode
- Generating thumbnails or previews from large PSD libraries
FAQ from Photoshop MCP Server
What are the system requirements?
Python 3.11 or later, macOS or Windows 10/11, and Adobe Photoshop 2023 or later.
Which backend should I use?
AppleScript (all versions on macOS), PowerShell (all versions on Windows), or UXP plug‑in (Photoshop CC 2021+ on both platforms, using WebSocket). Default depends on platform.
Does the server handle authentication?
No authentication or authorization is mentioned in the README. The server uses localhost by default.
What transport protocols does the server support?
REST (HTTP) and WebSocket endpoints are available, including streaming for thumbnail generation.
Where does my data live?
All file operations happen on the local machine where the server and Photoshop are running. The server does not upload data to external services unless using the LLM auto retouch feature (which calls an external LLM via LiteLLM).
