AI coding tools handle small, well-described fixes well: a button that overlaps the footer on mobile, a typo in a header, a console error on one page. That's most of what a review round produces.
The slow part is getting the feedback to them. Opening forty tasks one by one, copying the title, page, selector and console output, then saving each screenshot can take longer than some of the fixes.
pinreview's export does that step for you. One click downloads the whole board as a single file you can give to Claude Code, Cursor, ChatGPT or any other assistant.
What you get
Click Export on the board toolbar. pinreview downloads a ZIP named after the project and the date:
pinreview-acme-2026-09-14.zip
├── ISSUES.md
└── screenshots/
├── 12.png
└── 15.png
ISSUES.md starts with a summary table — number, title, severity, status and page for every task — then one section per board column, in board order. Each task carries what was captured when it was reported:
- severity, status, tags, assignee and due date
- the reporter and the page URL
- the CSS selector of the pinned element, plus fallback selectors
- the browser, viewport size and pixel density
- the description, console errors and comment thread
- links to any attachments
- the screenshot, embedded from the
screenshots/folder
The screenshots are files inside the ZIP, not web links, so the assistant can open them without network access or a login.
Here's one task from a typical export:
### #12 — Signup button overlaps footer
- Severity: High
- Status: To Do
- Tags: mobile, css
- Page: https://acme.com/signup
- Element: `main > form button.btn-primary`
- Browser: Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 …) Safari/604.1
- Viewport: 390 × 844 @3x
**Description**
> The button sits on top of the footer links, can't tap them.
**Screenshot**
The file contains no instructions for the AI, only facts about the tasks. You decide what to ask for.
Using it with an agent in your editor
Agents that work inside your repository, such as Claude Code or Cursor, work best with the file on disk.
- Unzip the export into a folder in your project, for example
feedback/. - Add that folder to
.gitignore. The export contains reporter emails and internal comments, so keep it out of your repository history. - Ask for a narrow batch:
Read feedback/ISSUES.md. Fix the tasks under "To Do" with severity High or
Critical. Look at each task's screenshot before changing anything, and use its
Element selector to find the component. One commit per task, with the task
number in the commit message.
If your tool can't open images from disk, attach the screenshots for the batch you're working on.
The selector matters more than it seems. "The signup button" might match five components. main > form button.btn-primary narrows it to one place in the markup, which the agent can search for. If your site uses generated class names that don't appear in your source, the selector still shows the page structure, and the page URL tells the agent which route to open.
Using it with a chat assistant
For ChatGPT, Claude or Gemini in the browser, upload ISSUES.md with the screenshots for the tasks you want to discuss. A chat assistant can't see your code unless you paste it in, so it's better at diagnosis than at the fix itself: "what's likely causing this console error?" or "what CSS would make this overlap happen at 390px wide?"
Skip the ZIP: connect directly with MCP
An export gives the assistant a snapshot. If your tool supports MCP (Model Context Protocol), as Claude Code, Cursor and VS Code do, you can connect it to pinreview instead and let it work with the live board:
- list tasks by status, severity, page or assignee
- open any task with its screenshot, selector and console errors
- comment on a task and move it to Done once it's fixed
Create a token under Settings → AI tools in pinreview; the page gives you the exact setup for your tool. The full walkthrough is in how to connect Claude Code or Cursor to your website feedback.
Tips for a good round
Triage first. The export includes every active task in every column, including Done, each labelled with its status. Archived tasks are left out. A quick triage pass — closing duplicates, setting severity, archiving opinions — turns the file from a raw pile into a clean work list.
Name the column. Tell the assistant which section to work on. "Fix everything in ISSUES.md" invites it to reopen items that are already done.
Keep batches small. Five related fixes you can review beat forty you can't. Group them by page or by component.
Check each fix against the screenshot. The assistant can read the report, but it can't see your staging site. Check each fix in a browser at the viewport size in the task, then move the card on the board so the reporter sees it was handled.
Export again for the next batch. An export is a snapshot. After you've worked through a batch and updated the board, export again for a fresh list.
Who can export, and what's included
- Export is on the Starter, Business and Agency plans. On Free, the button shows a Paid badge.
- Only project members can export. Guests reviewing through a shared link can't.
- Attachments are listed as links rather than copied into the ZIP, so a large video doesn't bloat it.
- Internal comments are included and labelled
(internal), since only your team can export. Keep that in mind before you share the file outside the team.
The bottom line
Your reviewers already captured the screenshot, selector, browser and console output. Export puts all of it into one file, so a whole review round can go to your AI coding assistant at once instead of task by task.
Handing the list to a person instead? Here's how to send your web developer a list of website changes. If your developers work in Jira, Trello or Asana, you can send feedback there with webhooks.