Website feedback MCP servers, compared

By LayerdockUpdated

Short answer

13 of the 15 website feedback tools we checked ship an official MCP server, so Claude Code, Cursor or ChatGPT can read feedback and act on it. They differ in what the agent gets: some pass the console log and network requests behind a bug, some only the comment and screenshot, and some cannot reply or close an item. Everything below was read off each vendor’s own MCP page on 19 November 2026. Layerdock is ours and is marked as such.

At a glance

ToolHow it connectsWhat the agent can readPlans
LayerdockoursLocal package @layerdock/mcp (npx), with a Layerdock tokenEach pin’s comment, CSS selector, screenshot, page URL, the console and network trace from the moment it was reported, and the element’s Inspect specs (spacing, radius, typography, colours)Pro and Agency, full context on both
BugHerdHosted (mcp.bugherd.com), OAuth, no API keyTask details: screenshot or recording, URL, CSS selector, browser and OS, severity, comments, attachments. Console and network logs not listedIn beta, free for all BugHerd subscribers
Marker.ioHosted (connect.marker.io), OAuth or access tokenIssues with context, screenshot, console log and network requests (where the plan captures them)Every plan at no extra cost; console and network data from the Team plan
UserbackHosted (mcp.userback.io), OAuthFeedback with comments and screenshots, console and network logs, searchTeam plan and up
UsersnapHosted (mcp.usersnap.com), OAuthSearch and filter feedback, single items, labels, topics with sentiment. Screenshots and console data not statedFree on every plan during the public beta
Jam.devHosted (mcp.jam.dev), sign-in or personal access tokenConsole logs, network requests, user events, screenshots, video transcript and frames, metadataListed on the Free plan
HuddlekitHosted (mcp.huddlekit.com), OAuthComments with element selector, page URL, viewport, thread and screenshot. Console and network logs not statedPro and Team plans
FeedbucketHosted (dashboard.feedbucket.app), OAuth or personal access tokenFeedback with screenshots, page, device, viewport, CSS selector, console logEvery paid plan, no extra cost
Simple CommenterHosted (OAuth) or local package @simple-commenter/mcp-serverComments with replies, element context and screenshots, search, reportsPaid plans and the 14-day trial; not the free plan
RuttlHosted, set up from workspace settings; OAuth or access tokenProjects, comments, tickets and threads. Screenshots and console data not statedPro plan
PageProoferHosted (app.pageproofer.com), bearer tokenPage, element selector and HTML, screenshot, browser and window size, console errors, failed network requestsEvery plan, including the trial
YbugHosted (mcp.ybug.io), OAuth; in betaPage URL, browser, device, screen size, screenshot and video links, console logs, network requestsPaid plans; read-only on Basic, changes from Startup
AtarimHosted (api.atarim.io), OAuthTasks, comments, sites, pages, screenshots, approval queue. Console data not statedFree to connect; some AI actions use credits
MarkUp.ioNone found on its own siten/an/a
Bird Eats BugNone found on its own site (an unofficial community package exists)n/an/a

Layerdock MCP (ours)

Connects
Local package @layerdock/mcp (npx), with a Layerdock token
Works with
Claude Code, Cursor, Codex, Windsurf, VS Code, Zed and any MCP client
Agent can read
Each pin’s comment, CSS selector, screenshot, page URL, the console and network trace from the moment it was reported, and the element’s Inspect specs (spacing, radius, typography, colours)
Agent can do
Reply on the thread, resolve the pin
Plans
Pro and Agency, full context on both
Source
/docs/connect-your-coding-agent

BugHerd MCP

Connects
Hosted (mcp.bugherd.com), OAuth, no API key
Works with
Claude Code, Claude, Cursor, ChatGPT, VS Code with Copilot, Windsurf, Codex CLI, Gemini CLI, Zed and others
Agent can read
Task details: screenshot or recording, URL, CSS selector, browser and OS, severity, comments, attachments. Console and network logs not listed
Agent can do
Update tasks (status, assignee, severity, due date), comment, create tasks, tags
Plans
In beta, free for all BugHerd subscribers
Source
https://bugherd.com/feature/mcp

Marker.io MCP

Connects
Hosted (connect.marker.io), OAuth or access token
Works with
Claude, Claude Code, Cursor, ChatGPT, VS Code, Windsurf, JetBrains, GitHub Copilot
Agent can read
Issues with context, screenshot, console log and network requests (where the plan captures them)
Agent can do
Status, assignee, priority, type, comments. No create-issue tool
Plans
Every plan at no extra cost; console and network data from the Team plan
Source
https://marker.io/marker-mcp

Userback MCP

Connects
Hosted (mcp.userback.io), OAuth
Works with
Claude, Claude Code, Cursor, ChatGPT, VS Code, Windsurf, OpenCode
Agent can read
Feedback with comments and screenshots, console and network logs, search
Agent can do
Create feedback, comment, assign, change workflow status, update
Plans
Team plan and up
Source
https://docs.userback.io/docs/userback-mcp

Usersnap MCP

Connects
Hosted (mcp.usersnap.com), OAuth
Works with
Claude, ChatGPT, Cursor, Perplexity
Agent can read
Search and filter feedback, single items, labels, topics with sentiment. Screenshots and console data not stated
Agent can do
Create opportunities and link feedback to them only; no replies or status changes
Plans
Free on every plan during the public beta
Source
https://usersnap.com/integrations/mcp

Jam.dev MCP

Connects
Hosted (mcp.jam.dev), sign-in or personal access token
Works with
Claude, Claude Code, ChatGPT, Cursor, Codex, VS Code, Gemini CLI, Copilot CLI, OpenCode
Agent can read
Console logs, network requests, user events, screenshots, video transcript and frames, metadata
Agent can do
Comments, reactions, update or delete a Jam, folders. No status or assignee
Plans
Listed on the Free plan
Source
https://jam.dev/mcp

Huddlekit MCP

Connects
Hosted (mcp.huddlekit.com), OAuth
Works with
Claude Code, Claude, Cursor and any MCP client over HTTP with OAuth
Agent can read
Comments with element selector, page URL, viewport, thread and screenshot. Console and network logs not stated
Agent can do
Status, priority, resolve, reply, tags, assignees. Cannot create or delete comments
Plans
Pro and Team plans
Source
https://huddlekit.com/mcp

Feedbucket MCP

Connects
Hosted (dashboard.feedbucket.app), OAuth or personal access token
Works with
Claude Code, Claude, Cursor, VS Code, ChatGPT, Codex
Agent can read
Feedback with screenshots, page, device, viewport, CSS selector, console log
Agent can do
Comments, update or resolve feedback, attachments, projects
Plans
Every paid plan, no extra cost
Source
https://feedbucket.app/features/mcp

Simple Commenter MCP

Connects
Hosted (OAuth) or local package @simple-commenter/mcp-server
Works with
Claude, Claude Code, Cursor, ChatGPT
Agent can read
Comments with replies, element context and screenshots, search, reports
Agent can do
Create and reply to comments, status, priority, tags, archive, plus project and team settings
Plans
Paid plans and the 14-day trial; not the free plan
Source
https://www.simplecommenter.com/docs/integrations/ai-agent

Ruttl MCP

Connects
Hosted, set up from workspace settings; OAuth or access token
Works with
Claude, ChatGPT, Cursor, Codex, Antigravity
Agent can read
Projects, comments, tickets and threads. Screenshots and console data not stated
Agent can do
Resolve threads; more not stated
Plans
Pro plan
Source
https://www.ruttl.com/mcp

PageProofer MCP

Connects
Hosted (app.pageproofer.com), bearer token
Works with
Claude Code, Cursor and any MCP client over HTTP
Agent can read
Page, element selector and HTML, screenshot, browser and window size, console errors, failed network requests
Agent can do
Comment, status, priority, assignee, due date. Moves notes to Review rather than closing them
Plans
Every plan, including the trial
Source
https://pageproofer.com/ai/

Ybug MCP

Connects
Hosted (mcp.ybug.io), OAuth; in beta
Works with
Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code, Lovable
Agent can read
Page URL, browser, device, screen size, screenshot and video links, console logs, network requests
Agent can do
Status, priority, type, assignee, tags, internal comments
Plans
Paid plans; read-only on Basic, changes from Startup
Source
https://ybug.io/features/mcp-server

Atarim MCP

Connects
Hosted (api.atarim.io), OAuth
Works with
Claude Code, Claude, Codex, ChatGPT, VS Code and any MCP client
Agent can read
Tasks, comments, sites, pages, screenshots, approval queue. Console data not stated
Agent can do
Create and update tasks, status, assign, comment; with its plugin, WordPress changes through an approval queue
Plans
Free to connect; some AI actions use credits
Source
https://atarim.io/mcp/

MarkUp.io MCP

None found on its own site. Checked on its own site on 19 November 2026.

Bird Eats Bug MCP

None found on its own site (an unofficial community package exists). Checked on its own site on 19 November 2026.

Common questions

Which website feedback tools have an MCP server?

Layerdock, BugHerd, Marker.io, Userback, Usersnap, Jam.dev, Huddlekit, Feedbucket, Simple Commenter, Ruttl, PageProofer, Ybug, Atarim. We found no MCP server on MarkUp.io’s or Bird Eats Bug’s own sites (checked 19 November 2026).

Which MCP servers give an AI agent console logs and network requests?

Layerdock, Marker.io (on plans that capture them), Userback, Jam.dev, PageProofer, Ybug and Feedbucket (console log) state it on their own pages. BugHerd, Huddlekit, Usersnap, Ruttl and Atarim do not list it.

Which ones let the agent close the loop by replying and resolving?

Layerdock, BugHerd, Marker.io, Userback, Huddlekit, Feedbucket, Simple Commenter, Ybug and Atarim can change status or reply. Usersnap’s server cannot reply or change status, Jam.dev’s has no status tool, and PageProofer moves a note to Review rather than closing it.

Is there a free website feedback MCP server?

BugHerd’s is free for subscribers during its beta, Usersnap’s is free on every plan during its beta (including the free plan), and Jam.dev lists MCP on its Free plan. Layerdock’s MCP is on the Pro and Agency plans.

Choosing a feedback tool for your coding agent rather than reading specs? See visual feedback tools for Claude Code and Cursor, the Layerdock MCP server, or how to connect your coding agent.