Clipacanvas

Clip.A.Canvas MCP Server

A universal MCP (Model Context Protocol) server that exposes Clip.A.Canvas HTML-to-video rendering as an MCP tool. Works with Claude Desktop, ChatGPT, Antigravity CLI (agy), Claude Code, and any STDIO/HTTPS-based MCP client.

Features

• Universal MCP tool — render_video and render_video_to_file exposed via the MCP protocol
• Works everywhere — Claude Desktop, ChatGPT, Antigravity CLI, Claude Code, and more
• Local rendering — no upload, no cloud; Chromium + FFmpeg run on your machine
• Base64 or file output — return video as base64 string or save directly to a path

Installation

Client Configuration

1. Antigravity CLI (agy)

Add the following to your mcp_config.json (typically at ~/.gemini/antigravity-cli/mcp_config.json):

{  "mcpServers": {    "clipacanvas": {      "command": "uvx",      "args": ["--from", "clipacanvas-mcp", "clipmcp"]    }  }}

Verify the server loads successfully by running agy inspect.

2. Claude Code

claude mcp add -s user clipacanvas -- cmd /c uvx --from clipacanvas-mcp clipmcp

3. Codex CLI

codex mcp add clipacanvas -- uvx --from clipacanvas-mcp clipmcp

[!NOTE] If codex mcp list displays Auth: Unsupported for Clip.A.Canvas, this is completely normal! This status simply means the server does not require external OAuth authentication to run (which is correct for local stdio MCP servers).

From PyPI

pip install clipacanvas-mcp# oruv tool install clipacanvas-mcp  # gives 'clipacanvas-mcp' command

Configuration

Environment variables (all optional):

MCP Tools

render_video

Read-only hint: No | Destructive hint: No

Render HTML/CSS/JS to MP4, returned as base64.

Arguments:

{  "code": "<!DOCTYPE html><html>...</html>",  "width": 540,  "height": 960,  "bitrate": "5M",  "frame_rate": 60,  "max_duration": 12}

Returns: Text content with video metadata and base64-encoded MP4 data.

render_video_to_file

Read-only hint: No | Destructive hint: Yes (overwrites file)

Same as above but saves the MP4 directly to output_path.

Arguments:

{  "code": "<!DOCTYPE html><html>...</html>",  "output_path": "/absolute/path/to/video.mp4",  "width": 540,  "height": 960}

Usage in AI Clients

Claude Desktop

Add to your claude_desktop_config.json:

{  "mcpServers": {    "clipacanvas": {      "command": "uvx",      "args": ["clipacanvas-mcp"]    }  }}

ChatGPT (via Bridge)

To use this local tool with ChatGPT, you can use an MCP-to-HTTPS bridge (like mcp-proxy) or host it on a public URL. ChatGPT currently requires an HTTPS endpoint for official "App" integrations.

Other Clients (Codex, Qwen Coder, etc.)

Since this is a standard STDIO-based server, it works with any client that supports MCP. Use uvx clipacanvas-mcp or python -m clipacanvas_mcp.server as the launch command.

Privacy Policy

Clip.A.Canvas is a local-first application.

• Data Collection: No user data, code snippets, or rendered videos are ever uploaded to any server.
• Processing: All HTML/CSS/JS rendering happens inside a local Chromium instance on your machine.
• Storage: Rendered MP4 files are saved only to the directory you specify.
• Third-party Services: None.

Requirements

• Python 3.10+
• FFmpeg (installed system-wide, via imageio-ffmpeg, or set CLIPACANVAS_FFMPEG_EXE)
• Playwright Chromium: installed automatically on first render if missing.

License

MIT