ADB MCP is an open-source proof of concept that lets AI assistants control Adobe Creative Cloud applications through the Model Context Protocol. With it, an LLM such as Claude can operate Photoshop, Premiere Pro, InDesign, After Effects and Illustrator from natural language instructions. The project has about 728 stars on GitHub and was updated in 2026.
The project has three parts: an MCP server that exposes Adobe app functionality to AI clients, a Node-based command proxy, and plugins (UXP for newer apps, CEP for older ones) that run inside the Adobe applications and execute the commands. It has been tested with Claude Desktop and the OpenAI Agents SDK, and in principle works with any MCP-compatible client. Example workflows include creating a Photoshop composition from a prompt, generating step-by-step tutorials from a live example file, renaming layers consistently, or building a Premiere project pre-populated with clips, transitions and effects. The author notes it is a community proof of concept, not endorsed or supported by Adobe.
Setup is documented per application, reflecting the different plugin systems involved: Photoshop, Premiere Pro and InDesign use UXP plugins, while After Effects and Illustrator are driven through CEP extensions, with a shared proxy keeping the connection alive.
ADB MCP is free and open source under the MIT license. You will still need your own Adobe Creative Cloud applications and an AI client, but the bridge itself costs nothing.
It is for designers, video editors and creative developers curious about AI-driven automation inside Adobe tools. Beginners can lean on it as a conversational tutor for Photoshop, while power users can offload repetitive production tasks to an assistant.