Add visual feedback to your Codex app

Why Codex teams need visual feedback

• Codex acts on your repo, so it needs unambiguous instructions about what to change and where.
• Free-text bug reports leave the agent guessing which component is at fault.
• Patchrooms attaches a selector, screenshot, and goal to every report.
• Codex then edits the right code path instead of hunting through the tree.

How Patchrooms works with Codex

1. Embed Patchrooms on your preview

   Add the script to the build your team reviews.

2. Capture element-anchored reports

   Reviewers click to comment; Patchrooms records the selector and context.

3. Hand the report to Codex

   Export Markdown and include it in your Codex task, or read reports directly over the MCP server (available now).

Add it in one prompt

Paste a script tagAdd in one prompt

Ask Codex to wire Patchrooms in for you, paste this prompt:

Add the Patchrooms review widget. Insert this script tag into the HTML entry point before the closing </body> tag, exactly as written:

…then give it the snippet below. It loads on the live preview and turns on artifact-review mode: pinpoint comments anchored to the elements your reviewers click.

<script src="https://room.patchrooms.com/v1/patchrooms.js"
        data-project-key="pr_xxx"
        data-source="codex"
        data-mode="artifact-review"></script>

Or try it locally: no account, no key, no server

Reviewing a Codex artifact on your own machine? Run Patchrooms in local mode: no account, no project key, zero network calls. Pinpoint comments stay in your browser and export to Markdown you paste straight back into Codex.

<script src="https://room.patchrooms.com/v1/patchrooms.js"></script>
<script>
  Patchrooms.init({ storage: 'local', mode: 'artifact-review' });
</script>

Review workflow

Share the preview link with your team or client. Anyone clicks an element, types a comment or records a voice note (Patchrooms auto-transcribes it to clean text), and captures the selector, a screenshot, and the page context. Every comment carries a status (new, triaged, in-progress, closed) so nothing falls through the cracks. See the full UI review workflow.

Export feedback to your AI agent

Each report exports as clean Markdown with the element selector, a screenshot reference, and the artifact goal and constraints. Paste it straight into Codex and the agent fixes the exact issue: this is the heart of AI artifact review. The Patchrooms MCP server lets agents read reports directly (list_reports / get_report) and triage them (set_status). Available now; extended MCP tools (create_room, issue creation) are planned, and a @patchrooms/react package is coming for tighter React integration.