cursor-feedback-extension

Cursor Feedback provides a Model Context Protocol (MCP) tool for interactive feedback within Cursor. The MCP server exposes an interactive_feedback tool that the AI agent can call to create a feedback request, which is then surfaced to you in the Cursor extension's sidebar. You can supply text, images, and file paths as feedback, and the AI will wait for your input before proceeding. The workflow enables a human-in-the-loop experience, where the AI can summarize work, display results in the UI, and request further guidance through the same MCP channel. To use it, configure the MCP server in your Cursor settings (or install via the one-click option) and restart Cursor. Then trigger the interactive_feedback process from the chat; the AI will present a summary or prompt for feedback in the sidebar, and you can provide the necessary feedback through the UI.

Prerequisites: Node.js installed, npm available, and Cursor installed. Follow these steps to install and configure the MCP server for Cursor Feedback.

1. Install the MCP server (one-click recommended) or install manually:

• One-click: Use the provided button in the README to configure the MCP server automatically.
• Manual configuration:

2. Manual installation (recommended for local control):

# Option A: One-click (recommended)
# Use the install button in the README (no code required here in CLI)

3. Manual configuration via npm (local install):

# Install the MCP package locally
npx -y cursor-feedback@latest

4. If you prefer a global setup, you can install globally and configure the MCP server:

npm install -g cursor-feedback

Then add to Cursor MCP config:

{
  "mcpServers": {
    "cursor-feedback": {
      "command": "cursor-feedback-mcp"
    }
  }
}

5. Restart Cursor to apply the configuration. The Cursor Feedback icon should appear in the sidebar once the MCP server is active.

Tips and caveats:

• The MCP server exposes the interactive_feedback tool. Use it to create a feedback session, after which the AI will wait for your input via the Cursor sidebar.
• You can customize timeouts and retry behavior with environment variables: MCP_FEEDBACK_TIMEOUT (seconds) and MCP_AUTO_RETRY (true/false).
• Ensure your Cursor extension supports the MCP tool calls and that the server is reachable by the extension's HTTP API.
• If you encounter issues with installation, verify that npm can install the package and that the shell can run npx cursor-feedback@latest without network restrictions.
• This MCP server supports multi-window/project isolation, image and file sharing, and Markdown rendering in summaries.