Bug reports your coding agent can fix, sent by the people who use your app
Your users report bugs from a button inside your product, and your code reports the errors they hit. Your coding agent reads both over MCP, fixes them and links the PR. The person who reported it gets an email once the fix is live.
- Remote MCP with OAuth, no API keys
- Works with any MCP client
- Public replies wait for your approval
The loop
From a user's report back to their inbox
You review the PR, deploy it and approve any public reply. The rest is the agent and Feedback.
A user reports it
From the button in your app, with a screenshot they can mark up, or from your public portal. While they type, similar posts show up so they vote instead of filing a duplicate.
Your agent picks it up
list_issues with kind BUG, sorted by votes, MRR or date. get_issue returns a Markdown brief with the report, what other users added, your team's notes and the links.
It fixes it and links the PR
claim_issue moves the post to in progress so no other agent grabs it. The agent fixes it with a test in your repo and calls link_pull_request.
The reporter hears back
Once the fix is in production, resolve_issue (or register_release with the post numbers) moves it to Shipped and everyone following it gets an email.
What the agent reads
Two kinds of reports, one brief format
get_issue builds the brief from structured data. Anything a user wrote arrives inside <untrusted_user_content>, so the agent treats it as data and never as instructions.
Bug reported by a user
- Title and description, in the user's own words
- Comments other users left on the same post
- Votes, plus the MRR of the customers who voted if your app sends it
- Your team's internal notes and the PRs already linked
- Board, status, tags and who has it
Error captured by your code
- Error type, message and likely source
- Stack trace from the latest event
- The last steps before it: navigation, clicks, console errors and warnings, failed requests
- Page, browser, release and any context you passed
- Occurrences, affected users, first and last seen, and whether it came back after a fix
- What users wrote when they chose to report it
Before anything leaves the browser, the SDK replaces values under keys such as password, token, cookie or card.
Other options
Other tools do this too
Several bug reporting tools already ship MCP servers. This is what each one says it does on its own site, checked on October 4, 2026. If one of them fits you better, use it.
| Tool | Who reports | What the agent gets | What the agent can do | Connection |
|---|---|---|---|---|
| Marker.ioSource | Your team, clients and site visitors through a website widget, no account needed | Screenshots, console logs, network requests, browser and OS | 33 tools: update status, priority, type and assignee, and comment | OAuth 2.1 with PKCE, or an access token |
| JamSource | Your team with the extension or Mac app, and users through Recording Links | Console logs, network requests, user events, screenshots and video | Mostly reads; can also rename, move, delete and comment on Jams | Remote server with OAuth |
| BugHerdSource | Clients pointing and clicking on your website | Screenshot, page URL, CSS selector, browser and OS | Create and update tasks, change status, comment | OAuth |
| UsersnapSource | Users through in-app feedback widgets | Feedback items, with semantic search and filters | Reads feedback and creates opportunities; can't edit raw items | OAuth 2.1 with PKCE (public beta) |
| MarkLayerSource | Anyone annotating a page, no account | Annotations with CSS selector, styles and the detected component | Acknowledge, resolve, dismiss and reply, live | Room URL with no sign-in, or local with npx. Free and open source |
| Melonhelp FeedbackSource | Your end users through the in-app button and public portal, plus errors your code captures | A brief per post: report, comments, votes and MRR; for errors, stack, last steps, page, browser and release | Claim, note, link the PR, resolve in a release, register the deploy, draft replies and the changelog | OAuth 2.1 with PKCE and dynamic client registration |
Marker.io
Source- Who reports
- Your team, clients and site visitors through a website widget, no account needed
- What the agent gets
- Screenshots, console logs, network requests, browser and OS
- What the agent can do
- 33 tools: update status, priority, type and assignee, and comment
- Connection
- OAuth 2.1 with PKCE, or an access token
Jam
Source- Who reports
- Your team with the extension or Mac app, and users through Recording Links
- What the agent gets
- Console logs, network requests, user events, screenshots and video
- What the agent can do
- Mostly reads; can also rename, move, delete and comment on Jams
- Connection
- Remote server with OAuth
BugHerd
Source- Who reports
- Clients pointing and clicking on your website
- What the agent gets
- Screenshot, page URL, CSS selector, browser and OS
- What the agent can do
- Create and update tasks, change status, comment
- Connection
- OAuth
Usersnap
Source- Who reports
- Users through in-app feedback widgets
- What the agent gets
- Feedback items, with semantic search and filters
- What the agent can do
- Reads feedback and creates opportunities; can't edit raw items
- Connection
- OAuth 2.1 with PKCE (public beta)
MarkLayer
Source- Who reports
- Anyone annotating a page, no account
- What the agent gets
- Annotations with CSS selector, styles and the detected component
- What the agent can do
- Acknowledge, resolve, dismiss and reply, live
- Connection
- Room URL with no sign-in, or local with npx. Free and open source
Melonhelp Feedback
Source- Who reports
- Your end users through the in-app button and public portal, plus errors your code captures
- What the agent gets
- A brief per post: report, comments, votes and MRR; for errors, stack, last steps, page, browser and release
- What the agent can do
- Claim, note, link the PR, resolve in a release, register the deploy, draft replies and the changelog
- Connection
- OAuth 2.1 with PKCE and dynamic client registration
Checked on each vendor's site on October 4, 2026. Features change, so check before you pick.
Where we differ
Built around the people who use your product
What Melonhelp Feedback adds, limited to what it does today.
Reports from real users
The in-app button opens as a bubble, a text button, a side tab or from a button of yours, and it's there for everyone who uses your product. Sign a token on your server and each report is tied to a real person, with their plan and MRR if you send them.
Next to your help desk
Feedback is part of the Melonhelp suite, alongside Melonhelp Tickets, the help desk for WhatsApp, email and Slack. One Melonhelp account signs you in to both. Each product is sold on its own.
Paste a URL, approve, done
A remote MCP server with OAuth. Claude Code, Codex, Cursor or any client that supports remote MCP connects by pasting the URL and approving in the browser. Teammates who should only read get a read-only URL.
The loop ends with an email
claim_issue, add_note, link_pull_request, resolve_issue, register_release, reply_to_user and draft_changelog. The email to the person who reported it goes out once the fix is in production.
Fair warning
When to pick something else, and what the agent can't do
Another tool fits better if
- You need session replay or a video of the bug. Jam records it.
- You want console and network logs on every report typed by hand. Marker.io and Jam attach them. In Feedback, those last steps come with the errors your code captures.
- Your reviewers are clients pinning comments on page elements. BugHerd and MarkLayer record the exact element.
- You want something free and open source to host yourself. MarkLayer is.
Guardrails on the agent
- It sees what your account sees, and only writes if you're an owner or admin.
- It can't delete anything, see secrets or touch billing.
- Replies it drafts wait for your approval, and the changelog stays a draft.
- Emails only go to verified addresses: a signed token from your app or a magic link.
- Release tracking depends on your plan. Without it, resolve_issue still sends the emails, right when you call it.
Point your agent at real user reports
Create your portal, add the button to your app and connect your coding agent.
Sign in with email, Google or Microsoft.