I make all my flyers and ads with an AI agent now. The annoying part was watching it screenshot headless Chrome over and over to find out the headline didn't fit at 728×90 😅
So I built keyline: an MCP server where the agent writes the design once as JSON, and every edit replies with what's broken at each size:
leaderboard content !overflow needs 572×116
It fixes the layout from the numbers, and only looks at a render to judge the taste. In a benchmark that took 2× fewer tokens than HTML + headless Chrome, and 6× fewer than Playwright MCP.
- one Rust binary on Skia, no browser
- PNG, JPEG, WebP, vector PDF, GIF, MP4
- Claude Code, Claude Desktop, Codex, Cursor or any MCP client
https://keyline.dev
What do you use for images in your agent workflows right now? Curious if anyone else hit the screenshot loop 👀
keylinekeyline: MCP server that designs images and video for AI agentsFree MCP server for Claude Code, Cursor and Codex: your agent designs social posts, ads, banners and video at every size from one JSON scene. No headless Chrome.
commit 5e3623d