From client note to fixed code: PageProofer and your AI agent

Website feedback has always had a gap in the middle. A client spots a problem and leaves a note. Then someone has to read it, work out which page and which element they meant, open the code, find the right file, fix it, and go back to tell the client it’s done.

PageProofer already closed the first half of that gap: every note is pinned to the exact element, with a screenshot, the browser, the window size and any console errors. Now your AI coding agent can close the second half.

PageProofer has an MCP server, the standard way for AI agents like Claude Code and Cursor to connect to other tools. Once it’s connected, your agent can read your notes, fix the code and report back on the note, all from one plain-language request.

Setting it up takes about a minute

1. In PageProofer, go to My Settings, open the AI agents tab and create a token.

2. Copy the command shown there (your token is already filled in) and run it in your project folder. For Claude Code it looks like this:

   claude mcp add --transport http pageproofer https://app.pageproofer.com/mcp --header "Authorization: Bearer YOUR_TOKEN"

   For Cursor and other agents, add an MCP server with the URL https://app.pageproofer.com/mcp and the header Authorization: Bearer YOUR_TOKEN.

3. Start your agent in the website’s project folder and ask it about your notes.

That’s the whole setup. No plugin, nothing to host, and it’s included on every plan, including during your free trial.

What it looks like in practice

Say a client leaves a note on the Northwind Coffee site: “The Add to cart button covers the price on my phone.” It’s note 12, pinned to the button on the product page.

You open your agent in the Northwind project and type:

“What’s wrong in note 12?”

The agent reads the whole note: the client’s words, the page address, the CSS selector and HTML of the button it was pinned to, and the fact that it was left in Safari on an iPhone with a 390px wide window. It looks at the screenshot too. In a few seconds it tells you the button overlaps the price on narrow screens, on the product page, and that the selector points at the product card’s footer.

“Find the code and fix it.”

The selector and HTML snippet are the best clues an agent can get. It searches the project, finds the product card styles, sees the footer only switches to a stacked layout above 480px, and fixes the breakpoint. It shows you the change before you commit.

“Commit that, comment on note 12 with what you changed, and move it to Review.”

The agent commits, posts a comment on the note saying which file changed and the commit hash, and moves the note to Review. Your client gets the comment by email, just like any other reply. Someone on your team checks the fix on the site and closes the note.

That’s three short messages for a fix that used to mean a round of emails, a hunt through the code and a status update.

Prompts worth stealing

There’s no special syntax. Ask the way you’d ask a teammate and name the site the first time. A few that work well:

Find the right notes

“What’s open on the Northwind Coffee site?”

“Show me the open notes tagged #checkout.”

“List the overdue notes, oldest due date first.”

Understand a note

“Look at the screenshot on note 7 and describe what’s broken.”

“Note 7 has console errors. What do they point to?”

Plan before you fix

“Find the code behind note 12 and tell me what you’d change. Don’t edit anything yet.”

Report back

“Add an internal comment to note 5 saying this needs a design decision first.”

“Ask the reporter on note 9 which browser they were using.”

Hand it the whole job

“Work through the open #mobile notes on Northwind Coffee. Fix each one, commit, comment with what you changed and move it to Review.”

There are more in How to use the AI agent.

You stay in charge

The agent acts as you, sees the same sites and notes you do, and everything it does shows in the note’s Activity tab, marked “via AI agent”. It can comment and change a note’s status, priority, assignee and due date, but it can’t mark a note Complete or Closed. Fixed notes land in Review, so filtering the dashboard by Review gives you a ready-made checklist of everything the agent worked on.

PageProofer never sees your code. The agent runs with your code on your computer; PageProofer just gives it the notes and saves the comments and changes it makes.

Better feedback, better fixes

The more a note carries, the better the agent does. Pin notes to the element rather than the whole page, add a screenshot when something looks wrong, tag related notes so the agent can work through them as a batch, and leave console capture switched on.

Ready to try it? See how it works, or start your free trial and connect your agent today.