Opens your agent, ready to go
The prompt is already typed. You read it and press Enter. Claude Code starts in your copy of the project’s GitHub repository.
Agent handoff
Pick comments and send them to your coding agent, with the page, element and source file. You stay in control of the change.
Your team picks comments in the widget’s inbox and sends them in one prompt. Cursor, Claude Code or Codex opens with the comments, page, element and source file filled in.
The prompt is already typed. You read it and press Enter. Claude Code starts in your copy of the project’s GitHub repository.
Every send also puts the whole prompt on your clipboard, so you can paste it into any other agent.
Choose Copy for AI agent on any comment, in the widget or the dashboard. Paste it into Codex, Cursor, Claude Code or another agent.
The page, branch, element, source file, browser and screenshot travel with the words.
CK-12 · /contact · feat/pricing button.contact-link · src/Contact.tsx:18 “A little more space around this?” Screenshot · Safari 18 · 1440 × 900
See exactly what the brief contains before it leaves CommentKit.
Copying doesn’t send a request to an agent or change your code. Your agent’s own permissions decide what it can do.
Codex, Cursor, Claude Code, VS Code or anything that takes text or speaks MCP.
Connect CommentKit as an MCP server. Your agent finds and reads feedback, including the screenshot, and replies when you allow it.
Add the address to your agent and sign in. Nothing to install.
Runs on your computer with a private agent token for one project. Needs Node.js 20 or newer.
The MCP prompt loads open feedback and its conversations. Your agent fixes it one by one, links the commit and marks it ready for review.
Agents read feedback and, if you allow it, reply and mark it ready for review. By default someone on your team approves each reply first. Resolving and deleting stay with people.