Feedback Loop

You set up and run the Feedback Loop MCP server to let humans review and guide AI-assisted development workflows. It provides a simple UI to run commands, see outputs, and send textual feedback back to the AI, improving collaboration and reducing unnecessary tool invocations.

How to use

You interact with the Feedback Loop MCP server through your MCP-compatible client. Start the server, then call the feedback_loop tool from your AI assistant to present a user-facing feedback UI. The UI lets you review command output, add comments, and return a structured feedback response to the assistant. When you need feedback, trigger the tool; when you’re ready to proceed, trigger it again or let the assistant continue based on the user’s input.

How to install

# Quick Start
npx feedback-loop-mcp

# Global installation
npm install -g feedback-loop-mcp
feedback-loop-mcp

# Local development setup
git clone <repository-url>
cd feedback-loop-mcp
npm install
npm run dev

Configuration

Configure the MCP client you use to connect to the server. The provided examples show how to wire the server into various clients.

{
  "mcpServers": {
    "feedback-loop-mcp": {
      "command": "npx",
      "args": ["feedback-loop-mcp"],
      "timeout": 600,
      "autoApprove": [
        "feedback_loop"
      ]
    }
  }
}

Cursor IDE

Add a dedicated MCP server entry to your Cursor configuration so the assistant can call the Feedback Loop tool when needed.

{
  "mcpServers": {
    "feedback-loop-mcp": {
      "command": "npx",
      "args": ["feedback-loop-mcp"],
      "timeout": 600,
      "autoApprove": [
        "feedback_loop"
      ]
    }
  }
}

Cline / Windsurf

Configure the server in your MCP settings to enable the Feedback Loop stream.

{
  "mcpServers": {
    "feedback-loop-mcp": {
      "command": "npx",
      "args": ["feedback-loop-mcp"]
    }
  }
}

Claude Desktop

Add the MCP server configuration to Claude Desktop so it can invoke the feedback loop during your AI-assisted workflow.

{
  "mcpServers": {
    "feedback-loop-mcp": {
      "command": "npx",
      "args": ["feedback-loop-mcp"]
    }
  }
}

Usage

The feedback_loop tool accepts a project directory, a descriptive prompt, and optional quick feedback options. You call it from an AI assistant to collect user feedback and then return the result to guide the next steps.

feedback_loop_mcp({
  project_directory: "/path/to/your/project",
  prompt: "I implemented the new user authentication flow with JWT. Does this approach meet your requirements?",
  quickFeedbackOptions: [
    "Yes, looks good!",
    "Needs minor changes to the error handling.",
    "Please try a different approach."
  ]
});

Running the Server

Choose how you want to run the server, then start it and verify it’s ready to receive feedback.

# Via npx (Recommended)
npx feedback-loop-mcp

# Via Global Installation
feedback-loop-mcp

# Local Development
npm start

Command Line Arguments

When running the server, you can supply optional parameters to predefine the project directory and initial prompt.

npm start -- --project-directory "/path/to/project" --prompt "Please review this code"

Prompt Engineering

Craft prompts that clearly request user feedback. Use the feedback_loop tool to present questions and collect input before proceeding.

Whenever you want to ask a question, always call the MCP feedback_loop tool.
Whenever you're about to complete a user request, call the MCP feedback_loop tool instead of simply ending the process.
Keep calling the feedback_loop tool until the user's feedback is empty, then end the request.

Tools

The server provides one primary tool for interaction.

feedback_loop

Troubleshooting

Common issues include the MCP server not connecting, missing npx, or permission errors on Unix systems. Ensure the server is running, Node.js and npm are installed, and the binary has execute permissions.

Development

For development and build information, a dedicated development guide covers setting up the environment and running in development mode.

License

MIT License - see package.json for details.

Available tools

feedback_loop

Displays a UI for collecting user feedback and returns the response