Tailr - Markup and iterate your projects live with your agent
Tailr is an app designed to streamline the process of iterating on projects with an AI agent, specifically addressing frustrations with making small UI changes and editing text. It allows users to mark up page elements and send them to their agent in batches, facilitating inline text editing and requesting multiple variations for specific elements. Tailr operates as an MCP server and proxy, injecting an overlay into the dev server, running locally with zero dependencies, and is MIT licensed for Node 18+ environments.
- Published
- Sep 7, 2026, 00:55
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Times shown in UTC
More details
I've been working with Claude Code for a bit now, and I've always found it especially frustrating to do two things: make many small changes to UI and edit its text. Funnily enough, these are two of the most important actions when building software with an LLM. Sweeping apps for AI-slop artifacts is very important, as an agent certainly won't catch every AI-tell, and AI-sounding copy on a landing page raises red flags for many people.
So, I built an app to help me build apps better. Tailr lets you mark up page elements and send them to your agent as a batch. It's an MCP server plus a proxy that injects the overlay into your dev server and runs completely locally with zero dependencies. MIT licensed; Node 18+. You can leave comments, edit text inline in its native formatting, and request multiple variations for a specific element. Or ask for a slider and the agent wires up a live parameter like a padding, glow, or duration that you scrub through on the page and keep wherever it looks right. Interacting with the app itself lets you dictate your design direction accurately rather than typing up a bulleted list in a textbox.
I've only been consistently using it in Claude Code and Cursor for Tailr's own landing page; curious how it holds up in other people's repos. If you could try it out (tell Claude to install it via the copy-paste prompt in the website/README since that's what needs to be stress-tested), let me know what you think — any feedback is greatly appreciated. Thanks!