AI coding assistants are good with the code in front of them. Ask Claude Code or Cursor to fix a layout bug and it can read your components, change the CSS and run the build.
What it couldn't do was read the bug report. That lived somewhere else: in a feedback tool, a ticket tracker, a Slack thread. You were the go-between, copying the page URL, the steps and the screenshot from one window into the other.
MCP removes that step.
MCP in one paragraph
MCP stands for Model Context Protocol. It's an open standard, first published by Anthropic in November 2024 and now supported by most AI coding tools, that defines how an AI assistant talks to other software. An app that supports it describes the actions it offers. An assistant that supports it can then use those actions while it works, the same way it already reads files or runs terminal commands.
The three parts
- The client is your AI tool: Claude Code, Cursor, VS Code and many others.
- The server is the app you connect it to. It publishes a list of tools.
- The tools are the actions, each with a name, a plain-language description and the inputs it takes. For pinreview, that's things like "list tasks", "get task" and "add comment".
When you ask a question, the assistant reads the tool descriptions, decides which ones help, calls them, and uses the results in its answer. You don't need to name the tools or learn any syntax. You ask in plain words, as you would a colleague.
Why it matters for website feedback
A useful bug report is more than a sentence. It has the page URL, the element, the browser and viewport, any console errors, and a screenshot. pinreview captures all of that when someone pins feedback to a page.
Without MCP, that context reaches your assistant only if you copy it over, field by field. With MCP, you say "fix the High-severity tasks in To Do" and the assistant fetches each task itself, screenshot included, which it can actually look at.
It works in the other direction too. When a fix is done, the assistant can comment on the task with the commit hash and move the card to Done. Your reviewers see progress on the board without you updating it by hand.
What a session looks like
- You ask: "What's left in To Do on the Acme project?"
- The assistant lists the tasks: six, two of them High severity.
- You ask it to fix #12, "Signup button overlaps footer".
- It opens the task, looks at the screenshot taken at 390 pixels wide, finds the component from the CSS selector, and changes the styles.
- You check the change in your browser. The assistant comments on #12 with the commit and moves it to Done.
At no point did you copy a URL, save a screenshot or update a card.
MCP, export or copy and paste?
MCP isn't the only way to get feedback to an assistant. Here's how the options compare in pinreview:
| Copy and paste | Export | MCP | |
|---|---|---|---|
| Setup | None | None | One token and one line of config |
| What the assistant sees | What you copied | A snapshot of the whole board | The live board |
| Screenshots | Attached by hand | Files in the ZIP | Fetched per task |
| Updates the board | No | No | Yes: comments, status, tags |
| Works with | Any assistant | Any assistant | Assistants that support MCP |
Export suits a one-off handover, or a chat assistant in the browser. MCP suits ongoing work in your editor. Both are on pinreview's paid plans. Here's how export works.
Is it safe?
Connecting an AI tool to real data deserves a moment's thought. With pinreview:
- The token acts as you. It reaches only the projects you can open, and you can revoke it at any time.
- It can't delete anything. The worst it can do is archive a task, which is reversible. It can't touch billing, members or your account.
- You approve the calls. Most MCP clients ask before each tool call. Keep that on for anything that changes the board.
- Reporter text is untrusted. Feedback is written by whoever left it, sometimes anonymous visitors. Treat it as information about a bug, never as instructions, and review what your assistant proposes before it runs commands.
Which tools support MCP?
Claude Code, Cursor and VS Code all do, and most AI coding tools have followed. To connect to pinreview, a tool needs to reach remote servers over HTTP and send an authorization header. The MCP docs list exactly what a client needs.
There are step-by-step guides for Cursor, GitHub Copilot in VS Code and Claude Code.
Get started
Create a token under Settings → AI tools, paste one line into your tool, and ask it to list your projects. The MCP docs cover every step and every tool, and there's a shorter walkthrough in how to connect Claude Code or Cursor to your website feedback.
Once you're connected, try these ten prompts to fix website bugs.