AgentCanvas

AgentCanvas is a hosted infinite canvas that headless AI agents control over MCP. You connect any MCP-speaking harness (Claude Code, Codex, Cursor, or a plain script) to its stdio MCP server, and the agent creates assets on a shared canvas that renders live over SSE. It supports three asset types - HTML docs in sandboxed iframes, images, and video - plus text-to-image generation via Fal. The MCP tool surface includes canvas_guide and get_design_system (so output comes out on-brand), create_canvas and list_canvases for multi-canvas work, create_html_asset, create_image_asset, create_video_asset, generate_image, and append_html / stream_html_demo for the live build effect where you watch docs assemble chunk by chunk. A connectors registry ("MCP of MCPs") brokers downstream research, image, and video models. Built on Next.js 16, React 19, and tldraw. It is a Developers Digest project.