AgentCanvas
A hosted infinite canvas that headless AI agents drive over MCP. Any MCP-speaking agent - Claude Code, Codex, Cursor, or a script - creates HTML docs, images, and video on a live canvas, streamed in as it builds.
Live canvas
Live embed of the running product. Open it in a new tab for the full experience.
How it works
Point any MCP-speaking harness - Claude Code, Codex, Cursor, or a plain script - at the stdio MCP server. No OAuth needed for local use.
The agent calls tools to place HTML docs in sandboxed iframes, images, and video onto a canvas, or generates images from text via Fal.
append_html and stream_html_demo push content in chunks over SSE, so you watch docs assemble live rather than popping in whole.
MCP tool surface
Every tool the MCP server exposes. Any harness that speaks MCP can call them to drive the canvas.
canvas_guideProgressive-disclosure skill: overview plus per-topic how-to for html, image, video, streaming, and research.
get_design_systemLoads the design system into context so every asset comes out on-brand.
list_canvases / create_canvasWork across multiple canvases, each with its own id and URL.
list_assetsList the assets currently on a canvas.
create_html_assetPlace a doc or slide inside a sandboxed iframe.
create_image_asset / create_video_assetPlace media on the canvas by URL.
generate_imageText-to-image via Fal, using the model from the connectors registry.
append_html / stream_html_demoStream content in chunks for the live build effect.
update_asset / delete_asset / clear_canvasEdit, remove, or reset assets on a canvas.
list_connectorsEnabled research, image, and video connectors - the "MCP of MCPs".
Get started
Connect Claude Code, Codex, Cursor, or a script over MCP and watch it build docs, images, and video onto a live infinite canvas.