MCP server

Every Feedbot install ships a Model Context Protocol server at /mcp/. Drop the URL into your MCP client and your AI assistant can read, triage, and reply to feedback.

Endpoint

• Cloud: https://app.feedbot.dev/mcp/
• Self-host: https://<your-domain>/mcp/ (or http://<server-ip>/mcp/ over plain HTTP)

Configuration

The same config file format works in Claude Code, Claude Desktop, Cursor, and Windsurf.

{
    "mcpServers": {
        "feedbot": {
            "url": "https://app.feedbot.dev/mcp/",
            "headers": {
                "Authorization": "Bearer fbk_live_..."
            }
        }
    }
}

Generate the API key from your project page (Project → API keys → Create), pick read or write scope based on whether you want the model to mutate state.

Available tools

• list_feedbacks — list with status / type / severity filters.
• get_feedback — fetch by ID.
• create_feedback — file new feedback.
• update_feedback_status — triage.
• add_note — annotate without changing status.
• reply_to_user — queue a reply that gets delivered through the original platform (Telegram / WhatsApp).

Use cases

• “Find the top 5 most-reported bugs from the last 7 days” — your assistant runs list_feedbacks(status=new, type=bug) and groups them.
• “Triage everything in acme/dashboard” — bulk sets status, severity, type from message content.
• “Reply to FB-A4B7D2 letting them know we shipped a fix” — drops a tracked reply that goes back to the original chat.