New: hand feedback to your AI agent over MCP

Hand website feedback straight to your AI agent

Connect Claude Code, Claude or Cursor to your workspace. Your agent reads each comment with its page, element and screenshot, fixes the code, then replies and resolves it.

Claude Code reading a Huddlekit comment over MCP, fixing the CSS, and the comment flipping to Resolved

From client comment to shipped fix with your agent doing the legwork

  1. A share link with guests joining one by one, no accounts
    Collect

    Clients comment on the live site

    Share a link and clients pin feedback on the exact spot. No account or extension needed.

  2. A comment's screenshot beside its captured page, OS, browser, size and selector
    Read

    Your agent gets the full picture

    It opens each comment with the element selector, page URL, viewport, reply thread and screenshot.

  3. A pinned page beside the code change an agent makes for it, then the pin resolves
    Resolve

    It fixes the code and closes the loop

    Your agent makes the change, replies on the thread and marks the comment resolved.

Skip the copy-paste between feedback and agent

Huddlekit MCP
Pasting into chat
Screenshot of the issue
Sent with the comment
Saved and pasted by hand
Element that was clicked
Exact selector
Described in words
Page URL and viewport
Captured automatically
Asked for and waited on
The discussion so far
Full reply thread
Summarised from memory
Replying to the reviewer
Posted on the thread
Written up separately
Status, priority, tags and assignees
Updated by the agent
Changed by hand afterwards
Finding what's left
Filter by status or category
Scrolling back through chat

Connect the AI tool you already use with one URL and no API key

Terminal
$ claude mcp add --transport http huddlekit https://mcp.huddlekit.com/mcp

$ claude
> /mcp
Add custom connector
Name
Huddlekit
URL
https://mcp.huddlekit.com/mcp
~/.cursor/mcp.json
{
  "mcpServers": {
    "huddlekit": {
      "url": "https://mcp.huddlekit.com/mcp"
    }
  }
}
Any MCP client
URL
https://mcp.huddlekit.com/mcp
Transport
Streamable HTTP
Auth
OAuth, found on its own

Your agent only gets what you grant and you can take it back any time

  • Your own sign-in

    No API keys. Each teammate approves their own connection and picks the one workspace it can see.

  • Read only by default

    Grant write access and the agent can also set status, reply, tag and assign.

  • Changes apply right away

    Make a connection read only or revoke it from the MCP page. It applies on the next request.

Connecting needs Pro or Team, and turning a tool off works on every plan. Read the setup guide

MCP is included on Pro and Team start free and upgrade when you connect

MonthlyYearly (save up to $78)
Free
$0
Free forever
  • 3 team members
  • 1 project
  • Unlimited guests
  • 100 MB storage
Pro
$16
Billed yearly ($190)
  • 3 team members
  • Unlimited projects
  • Unlimited guests
  • 5 GB storage
  • Connect Claude & Cursor (MCP)
TeamPopular
$33
Billed yearly ($390)
  • 15 team members
  • Unlimited projects
  • Unlimited guests
  • 50 GB storage
  • Connect Claude & Cursor (MCP)
  • Integrations
  • API access

Every limit and feature, plan by plan. Compare plans

Commonly asked questions and some more.

What is the Huddlekit MCP server?

It is a Model Context Protocol server at https://mcp.huddlekit.com/mcp. Connect Claude Code, Claude, or Cursor to your workspace and the agent can read your feedback and act on it: it opens a comment with its element selector, page URL, viewport, reply thread, and screenshot, then replies, retags, reassigns, or resolves it while it fixes the code.

Which AI tools can I connect?

The app walks you through Claude Code, Claude, and Cursor. Any other client that speaks MCP over streamable HTTP with OAuth works with the same server URL.

Do I need an API key?

No. Your AI tool signs you in with your normal Huddlekit account, and you pick the workspace and the access level on the approval screen.

What can a connected agent read and change?

With read access it can list the workspace's projects and comments, filtered by status or category, and open any comment in full. With write access it can also set a comment's status and priority, mark it resolved, reply on the thread, and apply tags and assignees. It can't create or delete comments, create new tags, or reach another workspace.

Can a comment trick my agent into doing something else?

Everything the server returns is marked as untrusted content written by users and site visitors, and the agent is told to report on it, never to follow it as instructions. You can also keep a connection read only, so it can never change anything in Huddlekit.

Do replies from my agent show up under my name?

Yes. Connections are personal, so a reply posted over MCP appears under your name and notifies people exactly like one typed in the app. Each teammate connects their own AI tool.

Can one connection cover several workspaces?

No. Each connection is pinned to the workspace you pick when you approve it. Connect again to work with another one.

Which plans include MCP?

Pro and Team. The MCP page is visible on every plan, so you can see the setup before upgrading. Removing write access or revoking a connection always works, and if a workspace moves to the Free plan its connections stop until you upgrade again. They also pause while a payment is past due. Nothing is deleted.

Have more questions? Visit our support center.

Collect feedback in minutes without the friction.