A website feedback tool lets your team or your clients click the spot on a website they mean, write a comment and send it with a screenshot attached. These tools are also called website annotation tools or client feedback tools. They replace emails with red arrows and lists like "the button on the second page". We compared five tools built for review rounds and QA before go-live: sitebrunch, Feedbucket, Pastel, ruttl and MarkUp.io. They differ in what the cheapest plan costs, how clients give feedback and what happens to the feedback afterwards. Four of the five can now pass comments to AI coding assistants such as Claude Code, Cursor or OpenAI Codex, which then change the code. In sitebrunch, comments, the AI connection and unlimited guests are free for one website.
Quick picks
Free, with AI and room to grow: sitebrunch. Comments and the MCP server are free for one website, tracking and accessibility checks for the same site come with the paid plans.
Clients comment without installing anything: Feedbucket, with many integrations on all plans.
Quick client reviews with a link: Pastel.
Small team on a tight budget: ruttl, from $15 per user.
Many reviewers, pay per review instead of per user: MarkUp.io.
Bug reports with console logs for testers: a bug tracker such as BugHerd or Marker.io, see the end of this article.
Prices and plans
Tool | Cheapest paid plan | Free plan | Company and data location |
|---|
sitebrunch | €9/month (Starter), 1 user, unlimited guests. Comments are free | Yes: 1 website, 1 user, unlimited guests, with comments, awork and MCP server | Germany (Hamburg), data in the EU |
Feedbucket | $39/month billed yearly, $49 monthly, 2 members, unlimited reporters | No, 14-day trial | Sweden, data in the EU (Frankfurt) |
Pastel | $29/month billed yearly, $35 monthly, 2 users | Yes: 1 user, 1 canvas | Canada, data location not published |
ruttl | $15/month per user billed yearly, $18 monthly | Yes: 1 project, up to 5 collaborators | India |
MarkUp.io | $79/month for 50 MarkUps, billed monthly | No, 14-day trial with card | USA |
How clients give feedback
Tool | How feedback is captured | Client needs an account | Client installs something | AI connection (MCP) |
|---|
sitebrunch | Browser extension or dashboard | Yes | Browser extension | All plans, also Free |
Feedbucket | Script or plugin on the site | No | No | All plans |
Pastel | Shared link | No | No | Plan not published |
ruttl | Shared link, extension for web apps | No | No | Pro plan |
MarkUp.io | Shared link, Chrome extension | No | No | None |
Sources: each vendor's pricing page, documentation and MCP documentation, checked on November 17, 2026. sitebrunch prices are in euros. The other vendors show prices in "$".
sitebrunch
Best for: teams and agencies that want to start for free and add tracking and accessibility for the same website later.
You comment with the sitebrunch browser extension directly on the website: press C, click the spot you mean and write. Every comment gets a screenshot with a timestamp. You tag the person responsible, who is notified by email, and turn the comment into a task in awork if you like. Through the MCP server, Claude Code, Cursor or Codex read the comments, fix them in the code and reply. For one website, all of this is free: comments, awork, the MCP server and unlimited guests. From the Starter plan, the same project also shows click heatmaps, conversions and daily accessibility checks for every page you comment on.
Limits: the extension runs in Chromium-based browsers such as Chrome, Edge and Brave. Guests sign in with a sitebrunch account. Free, Starter and Solo include one user account, the rest of the team joins projects as guests.
Feedbucket
Best for: agencies that want clients to comment without installing anything.
You add a script to the site, and clients comment without an extension or an account. Feedbucket connects to many tools on all plans, including Jira, GitHub, Asana, Trello and Slack. Its MCP server hands the AI assistant screenshots and the CSS selector of the element, on the Business plan also console logs.
Limits: no free plan, and every site needs the script.
Pastel
Best for: quick client reviews with a link.
Pastel shows your website through its own server, and you comment on top of it. Clients comment with a link and need no account. Through an MCP server or a command-line tool, AI agents read the comments with their screenshots, reply to them and resolve them.
Limits: integrations such as Jira and Asana start on the Team plan, Pro only offers a CSV export. Staging sites must be reachable by Pastel's servers.
ruttl
Best for: small teams on a tight budget.
The Pro plan starts at $15 per user a month. For one or two people that is cheap, for a bigger team the per-user price adds up. You paste a URL into a project, and clients comment with the project link without logging in. On the Pro plan, an MCP server gives AI agents the comments and threads to fix in your code.
Limits: the free plan covers 1 project, logged-in web apps need the Chrome extension and the MCP server needs the Pro plan.
MarkUp.io
Best for: teams with many reviewers that want to pay per review, not per user.
You paste a link and comment on the live site, and guests need no account. MarkUp.io sends comments to Slack, Microsoft Teams and Zapier and has an API. You pay by the number of MarkUps per month.
Limits: no Jira or Asana integration and no AI features. Password-protected sites can be tricky because the site loads inside the tool.
Collecting feedback is the easy part. The work starts afterwards: someone reads every comment, finds the right file, changes the code and tells the person who asked that it is done. For a relaunch with 80 comments, that can take days.
An AI coding assistant can take over a large part of that, but only if it knows where the feedback points. An email that says "the button on the pricing page looks off" is not enough. A comment pinned to the exact spot on the page is.
The tools connect through MCP (Model Context Protocol), an open standard that lets AI assistants use data from other tools. What the assistant receives differs:
Tool | What the assistant receives (per vendor) | Plans |
|---|
sitebrunch | Comment thread, CSS selectors, click position, page, screen size, device; it can reply, resolve and reopen | All, also Free |
Feedbucket | Screenshots as images, CSS selector, URL, browser, screen size, console logs on Business | All |
Pastel | Comments with replies and screenshots, the spot on the page the comment is pinned to; it can reply, resolve and assign | Not published |
ruttl | Projects, comments, tickets and threads; it can resolve threads | Pro |
MarkUp.io | No MCP server | |
What matters most:
The CSS selector tells the assistant which element is meant. With it, the assistant can search your code for the component that renders this element instead of guessing.
The thread carries the full discussion, so the assistant sees the latest decision, not just the first comment.
Write access decides whether the assistant can reply and close the comment, or whether you have to do that by hand.
What AI should not decide alone
AI works well for clear, small changes: a wrong text, a broken link, spacing, a color, a missing label. It works less well when a comment is a matter of taste ("this feels too busy") or when two comments contradict each other. Tell the assistant to reply to unclear comments with a question instead of guessing. And always review the code before it goes live. The assistant writes the change, but you are responsible for it.
Feedback rarely stands alone. During a review round, someone writes "nobody will find this button". Before a relaunch, you want to know how the old site performed. Before go-live, the new templates need an accessibility check. With separate tools, that means separate logins, separate scripts and separate exports.
In sitebrunch, all of this sits on the same page of the same project:
That helps most when the feedback is part of a bigger project. For a website relaunch, you record the baseline on the old site, collect feedback on the new one, check accessibility before go-live and compare the numbers afterwards, all in one tool.
Plan | What you get | Price |
|---|
Free | For one website: comments, awork, MCP server for comments, unlimited guests, automated accessibility check every 7 days | €0 |
Starter | Plus unlimited websites, metrics, live click heatmaps, journeys, conversions and goals, daily accessibility checks, MCP server for fixing accessibility issues | €9/month |
Solo | Plus manual accessibility audits (1 at a time) and export of findings to Jira, Notion or Asana | €49/month |
Agency | Plus unlimited users and 2 audits at a time | €149/month |
Details are on the pricing page.
How it works in sitebrunch
Collect comments. Your team or your clients comment on the live site or the staging site with the extension.
Connect your AI assistant. In Claude Code, run claude mcp add --transport http sitebrunch https://mcp.sitebrunch.com/. The first time you use a sitebrunch tool, your browser opens and you sign in once. The setup guides also cover Codex, Cursor, Windsurf, Claude Desktop and GitHub Copilot in VS Code.
Give the assistant a task. For example: "Go through the open sitebrunch comments on the home page, fix what can be fixed in the code, and reply to each one with what you changed."
Review the changes. The assistant changes the code in your project. You check the diff, test it and deploy it as usual.
Close the loop. The assistant replies to each comment and marks it as resolved. For a task, the reply also goes to awork. The person who left the feedback sees the answer in sitebrunch.
When you search for "website feedback tool", you also find tools that do a different job.
Bug trackers are built for testing teams. They record console logs and network requests and send bug reports straight to Jira:
BugHerd: extension or script, Jira, GitHub and Linear, MCP server in beta. From $42 a month billed yearly, data in the US.
Marker.io: records console logs, network requests and session replays, MCP server on all plans. From $39 a month billed yearly, Jira from the Team plan, data in the EU.
Userback: bug reports plus feature requests and surveys, free plan with feedback visible for 7 days, MCP server from the Team plan. From $39 a month billed yearly, data in the US.
Usersnap: now a platform for product teams that collect feedback from many channels. From €49 a month, data in the EU.
Survey tools such as Hotjar or Typeform ask visitors questions with a feedback button or a pop-up. They tell you what visitors think, not what your team wants to change on the page.
Frequently asked questions