OpenPencil — AI-Native Vector Design + MCP

Main

• Treat .op as design-as-code: keep it in Git, review diffs, and generate exports (React+Tailwind/HTML+CSS) from the same source file.

• Use MCP for automation: README calls out one-click MCP install into multiple AI CLIs so agents can read/modify .op files deterministically.

• Scale with concurrency: the README describes orchestrated “agent teams” working on different page regions in parallel.

• Start with docker web mode for evaluation, then move to the desktop app if you need native file associations and local workflows.

Source-backed notes

• README lists install options: Homebrew cask (macOS), Scoop (Windows), direct downloads, and a global CLI package.
• README states .op files are JSON and designed to be diffable and Git-friendly.
• README highlights a built-in MCP server for automating .op file reads/edits from MCP-compatible agents.

FAQ

• Is it just prompts-to-images?: No — README positions it as a vector design tool with design-as-code files and code exports.
• How do agents automate it?: Use the built-in MCP server; agents can read/create/modify .op files through MCP tools (per README).
• What’s a safe first run?: Start the web container (ghcr.io/...:latest) and generate a small page before scaling to multi-agent teams.