Triage product feedback in Cursor

Setup

You need a secret key (nv_live_…) from Dashboard → API keys.

{
  "mcpServers": {
    "nordva": {
      "url": "https://mcp.nordva.dev",
      "headers": {
        "Authorization": "Bearer nv_live_..."
      }
    }
  }
}

Then ask Cursor

• Show me the latest bug reports.
• How many feature requests came in this week, and what are they about?
• Log this as feedback from user 4812: the export button does nothing on Safari.
• Route bugs to our GitHub repo acme/app.

For “Show me the latest bug reports.” Cursor calls list_feedback with:

{
  "classification": "bug",
  "limit": 20
}

Returned: A list of submissions with id, text, classification, confidence and page URL, plus a cursor if there are more.

Feedback tools and arguments

Behaviour worth knowing

• Feedback works on every plan. Free takes 50 messages a month per project and sorts them by keywords; on Indie and Builder, AI sorts them and adds a one-line summary, and routing reaches Slack, Linear and GitHub as well as email.
• submit_feedback takes between 10 and 2,000 characters of text, plus an optional category_hint and a 1 to 5 rating. With AI sorting the classifier still decides and reports its own confidence; with keyword sorting (Free) the hint is the category.
• configure_feedback_routing replaces any existing configuration for the same destination. It creates and updates but does not delete.
• Routing credentials (a Linear API key, a GitHub token, a Slack webhook URL) pass through the agent. The tool description tells it to ask for consent first and never to repeat the secret back.
• Cursor's agent decides when to call a tool from the tool descriptions; naming the tool in the prompt ("use publish_changelog_entry") removes the guesswork.
• Tools that send email or remove content are flagged destructive by the server, and their descriptions tell the agent to confirm with you first.