Your users report, your agent fixes, your users hear back
A feedback portal, an in-app button and error reporting for the product you're building, wired to your coding agent over MCP. Start on the free plan.
- Free plan, no card
- Flat price, never per user or vote
- MCP with OAuth, no API keys
How it works
One loop, from report to fix to email
Reports, fixes and notifications in one place, without copy-pasting between tools.
Users report
Ideas and bug reports come in from a button in your app or from your public portal. Errors your code captures arrive grouped, with the page, the release and who hit them.
Your agent fixes
Claude Code, Codex, Cursor or any MCP client lists issues by affected users, votes, MRR or date, claims one, fixes it, links the PR and resolves it.
Users hear back
Whoever asked for it or reported the error gets an email when it's resolved. Resolve it in a release and the email waits until that release is deployed.
Free plan
Start on the free plan
It doesn't expire and doesn't ask for a card. The portal, the in-app button, error reporting and the MCP connector with read and write tools are all in it.
Free, US$0
- Public portal with boards, votes, roadmap and changelog
- In-app button with screenshots
- Error reporting, 5,000 events a month
- MCP connector with read and write tools
- Users identified with a token from your backend
- Unlimited end users and votes
- 2 boards, 2 admins and 1 survey
When you need more
- Indie, US$15/month
- Custom domain, no Melonhelp branding, boards for identified users only and release tracking. 10 boards, 5 admins, 5 surveys and 50,000 error events a month.
- Pro, US$39/month
- Multiple products or portals, unlimited boards, admins and surveys, and 500,000 error events a month.
Yearly billing brings them to US$12 and US$32 a month. Over the monthly error events, the SDK switches to sampling: nothing breaks and there's no surprise bill.
SDK
One script tag in your shared layout
Copy it with your public key from Settings, Feedback button, and allow your local and production origins.
- Your users get a feedback button with screenshots. It opens as a bubble, a text button, a side tab, from your own button or only from code.
- data-capture-errors catches uncaught errors, silently with auto or asking the user first with ask.
- data-release groups errors by version, so a fixed error that comes back reopens as a regression.
- Sign an HS256 JWT on your server to tie each report to a real user. They don't create another account, and emails only go to verified addresses.
- The panel follows the browser language (English, Spanish or Portuguese), or set it with data-lang.
<script
src="https://feedback.melonhelp.com/sdk.js"
data-key="pk_..."
data-release="1.4.2"
data-capture-errors="ask"
async
></script>try {
await saveOrder(order);
} catch (error) {
window.MelonFeedback("captureError", error, {
mode: "ask",
context: { orderId: order.id },
});
throw error;
}// token: HS256 JWT signed by your server
window.MelonFeedback("identify", { token });MCP and agent skill
Hook up your coding agent
The MCP server gives your agent access to your workspace. The skill teaches it how to integrate Feedback into your repo.
MCP server
Works with Claude Code, Codex, Cursor or any client that supports remote MCP. Paste the URL, a browser opens and you authorize with your Melonhelp account. No API keys, nothing to store.
https://feedback.melonhelp.com/mcpAgents that read their config from the repo can use a file such as .mcp.json:
{
"mcpServers": {
"melonhelp-feedback": {
"type": "http",
"url": "https://feedback.melonhelp.com/mcp"
}
}
}Read-only connection: https://feedback.melonhelp.com/mcp/readonly
Agent skill
Run it once in your repo root. It adds the melonhelp-feedback skill, built on the open Agent Skills standard, so it works with any compatible agent.
npx skills add https://feedback.melonhelp.comThen ask your agent:
Integrate Melonhelp Feedback into this project using the melonhelp-feedback skill and its MCP server.
Before writing code, clarify with me whatever you can't infer from the repository.It detects your framework, writes the code with your real key, allows your origins and checks that it loaded. The identity secret never goes through the agent: it leaves MELON_FEEDBACK_SECRET empty for you to fill in.
MCP tools
What your agent can do
Read tools work on any connection. Write tools show up on /mcp for owners and admins.
Set up
get_install_guide- Steps and exact code for your framework and backend, with your real key
check_installation- Which sites loaded the SDK, which were rejected, and whether users, requests and errors are arriving
add_allowed_originswrite- Allows your local and production origins
configure_launcherwrite- Bubble, button, tab, your own button or code only
create_surveywrite- Creates a satisfaction survey
Try asking
- Install Melonhelp Feedback in this project.
- Fix the three errors affecting the most users and link the PRs.
- We just deployed 1.4.3 to production, register it.
- Which requests have the most votes and are still open?
- Draft the changelog with what we resolved this week.
Work the queue
list_portals- The portals in your workspace
list_issues- Errors and requests by affected users, votes, MRR or date
get_issue- One case with the context to fix it
list_errors- Error groups with their occurrences and versions
claim_issuewrite- Claims it so two agents don't do the same work
add_notewrite- Leaves an internal note
link_pull_requestwrite- Links the PR or commit with the fix
resolve_issuewrite- Marks it resolved, optionally in the release that ships the fix
register_releasewrite- Registers a deployed release and notifies the people who reported what it fixes
reply_to_userwrite- Drafts a public reply that waits for your approval
draft_changelogwrite- Drafts the changelog entry
Clients that support MCP prompts also get install_feedback, add_error_reporting and add_survey as commands.
Guardrails
- The agent 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.
- Public replies wait for your approval and the changelog stays a draft.
- What your users write reaches the agent marked as untrusted content, kept apart from its instructions.
- Every change is logged under the person who connected it. End the session and access stops on the next call.
Release tracking
Close the loop on every release
Release tracking comes with Indie and Pro. On the free plan, resolving an error emails the people who reported it right away.
Claim it
The agent calls claim_issue, so two agents never pick up the same work.
Fix it and link the PR
It fixes the bug in your repo with a test and calls link_pull_request.
Resolve it in a release
resolve_issue with the version that ships the fix, for example 1.4.3.
Register the deploy
Once 1.4.3 is in production, register_release (or the Register deployed release button in Errors) emails the people who reported it. If the error shows up again in 1.4.3 or later, it reopens as a regression.
Docs
Read the details
Everything above, step by step.
Integrate into your app
With your coding agent or by hand: SDK, allowed origins, user identity, language, errors, releases, surveys and CSP.
MCP connector
Both URLs, roles, agent sessions, every tool, the ready-made prompts and security.
Melonhelp suite
Part of the Melonhelp suite
Feedback is one of the Melonhelp products, and one account signs you in to all of them. Tickets and Agents have their own MCP servers too. Each product is sold on its own.
Explore the suiteTry it on your own product
Create your portal, install the skill and let your agent wire it up.
Free plan, no card. Sign in with email, Google or Microsoft.