Skip to content

teseife/diffusion-studio

v0.1.3MIT

Create, edit, inspect, and preview local video projects with Diffusion Studio.

GitHub

Diffusion Studio

An independent Codex and Claude Code plugin for the Diffusion Studio desktop app. Version 0.1.3 adds Claude Code marketplace support. It includes the app's MCP connection, an editing skill, an original independent icon, and a read-only connection diagnostic. It does not bundle or modify Diffusion Studio.

Requirements

  • Codex or Claude Code with plugin support and local terminal/file access.
  • Diffusion Studio installed and running on the same computer as the client.
  • The app's dapi CLI, installed in Settings → MCP & CLI, for launching and shell workflows.
  • Python 3 only if you use the bundled diagnostic.

The connection uses the app's Streamable HTTP server at http://127.0.0.1:3274/mcp. It needs no additional authentication for this local connection. Generative and analysis features can use the services and account configured in Diffusion Studio; basic composition and local previews do not require this plugin to obtain an OpenAI API key.

Install in Codex from this marketplace

From the repository root:

codex plugin marketplace add .
codex plugin add diffusion-studio@diffusion-studio-community

Start a new Codex chat after installation so it can load the plugin's tools and skill. Keep Diffusion Studio running. Invoke Diffusion Studio and ask it to check the connection or create a short title card in a new project.

Install in Claude Code

From the repository root:

claude plugin marketplace add "$PWD"
claude plugin install diffusion-studio@diffusion-studio-community

Start a new Claude Code session. Verify the connection in /mcp, then ask for a video edit or invoke /diffusion-studio:diffusion-edit. See the Claude Code guide for GitHub and agent-assisted installation.

Connection diagnostic

Run the diagnostic from the plugin folder:

python3 scripts/doctor.py

It checks CLI availability, the MCP handshake, tool discovery, and a read-only context request. It does not open a project, generate media, export, or publish anything. A successful result has mcpReady: true. Missing dapi does not prevent an already-running MCP server from working.

Package layout

The root plugin.json and mcp.json follow the portable Agent Plugins schemas. .codex-plugin/plugin.json supplies OpenAI presentation and the compatibility MCP/skill declarations; .claude-plugin/plugin.json provides Claude Code metadata, which automatically discovers the same skills/ directory and .mcp.json HTTP configuration used by Codex. Keep both MCP files in sync. Skills read the documentation shipped with the installed Diffusion Studio version rather than bundling a potentially outdated copy.

Limits and publication

This version works with local Codex or Claude Code execution on the computer running the editor. A remote ChatGPT, Claude, or cloud coding environment cannot reach your computer's loopback address. The local HTTP server is not a public submission endpoint.

For architecture and known limitations, see the repository-root README. This distribution includes no official Diffusion Studio brand assets. This package has not been submitted to or approved by OpenAI or Anthropic and is not affiliated with Diffusion Studio.

References: Diffusion Studio, Codex plugin packaging, public submission.