infinite-canvas is an open‑source, node‑based workbench for image creation that combines multiple canvases, draggable nodes, connections, a mini‑map, undo/redo and import/export operations. The browser front‑end directly invokes any OpenAI‑compatible image generation API, enabling text‑to‑image, image‑to‑image, reference‑image editing, text Q&A and audio/video synthesis. A canvas helper provides conversational context around selected and upstream nodes, inserting generated results back into the workspace, while a local Canvas Agent uses the MCP protocol to control the canvas via Codex or Claude Code. The project ships a Codex App plugin that auto‑registers the MCP and launches the local agent, and a plugin system that loads remote node modules through URLs with a TypeScript SDK for custom node development. Users can define custom request scripts for image or video generation to adapt to third‑party endpoints. An internal prompt library contains seven open‑source prompt collections and allows custom JSON prompts cached in IndexedDB. Deployment is handled by Docker‑Compose (default port 3000) with a configuration page for the API base URL and key. The repository is MIT‑licensed and currently in early development, with no guaranteed data‑compatibility guarantees.