Open‑Source MCP Server for Adobe Creative Apps

Published: 30/July/2025 Views: 322

What is adb-mcp?

Adb‑mcp authored by Mike Chambers, is an open‑source, proof‑of‑concept MCP (Model Context Protocol) server that enables AI clients and agents to control Adobe Photoshop, Premiere Pro, and InDesign. It bridges the gap between large‑language models and native creative tools, empowering AI‑driven creative workflows.

  • Hosted on GitHub (mikechambers/adb‑mcp), licensed under MIT, with hundreds of stars and forks.
  • Provides functionality for three Adobe applications:
    • Photoshop — most comprehensive support
    • Premiere Pro — via beta UXP integration; basic functionality
    • InDesign — basic document creation; UXP support available
  • Tested with AI clients like Claude Desktop and the OpenAI Agent SDK. Should work with any MCP‑compliant client.

Architecture & Components

The architecture consists of three main parts:

  • Python MCP Server — exposes AI‑facing commands for Adobe tool functionality.
  • Node.js Proxy Server — relays MCP commands to the Adobe UXP plugin, since UXP plugins cannot act as socket servers.
  • Adobe UXP Plugin — installed into the Adobe app (Photoshop, Premiere, InDesign); listens for commands and executes actions.

Flow:

arduino

AI ⇄ MCP Server ⇄ Proxy Server ⇄ UXP Plugin ⇄ Adobe App

Called a proof‑of‑concept because it relies on this three‑part bridge rather than native socket support in Adobe I/O

Adb-mcp Use Cases & Examples

  • Generate and manipulate design files (social posts, double‑exposure composites) end‑to‑end via AI prompts.
  • Batch tasks such as renaming layers, organizing files, applying styles.
  • Automatically generate tutorials: AI creates a sample file and a step‑by‑step guide retraceable by a human user.
  • Build slideshows or music‑video sequences in Adobe Premiere using AI‑selected clips and effects.

Experiment examples include layer cleanup, tutorial generation, sequence assembly, content description, and more.

Requirements & Installation

Prerequisites:

  • Claude Desktop or other MCP‑capable AI clients (tested on Mac and Windows).
  • Python 3 and Node.js.
  • Adobe Creative Cloud with Adobe Photoshop (v26+) or Premiere Beta (v25.3 Build 46+).
  • Adobe UXP Developer Tool (for plugin installation).

Quick‑start setup:

  • Download latest release from GitHub: includes Claude MCP installers (.dxt), Adobe plugins (.ccx), and proxy executables.
  • Install Claude Desktop and load the .dxt file for Photoshop or Premiere.
  • Unzip and run the proxy (e.g. adb‑proxy‑socket‑macos‑x64 or …‑win.exe) to launch the websocket proxy.
  • Load the corresponding .ccx plugin in UXP Developer Tool, then launch the plugin UI panel in the Adobe app and click Connect.
  • In Claude Desktop, insert config://get_instructions resource to load guidance for the session.

Limitations & Areas for Improvement

  • Not production‑ready: setup is non‑trivial and requires proficiency with command‑line tools and MCP configurations.
  • Separate proxy server is needed because UXP plugins currently cannot act as listening sockets.
  • Premiere and InDesign support is currently limited compared to Photoshop.
  • No opportunity for native integration—Adobe would need to support MCP or allow plugins to open sockets directly.

Future Directions

  • Improving usability: simplified installer, better documentation, streamlined proxy configuration.
  • Expand application support, including Illustrator and other UXP‑based Adobe tools.
  • Deeper integration options: including direct MCP within apps or Creative Cloud platform.
  • Coupling with other MCPs (e.g. BlenderMCP, memory MCP servers) for richer workflows and enhanced automation.