MCP Feedback Server

An interactive feedback system for MCP (Model Context Protocol) that allows agents to request user feedback during task execution.

Overview

This system consists of:

• feedback_server.py: The main MCP server that runs in your terminal and handles feedback requests

• feedback_client.py: A test client script demonstrating how agents connect to request feedback

Related MCP server: MCP Feedback Collector

Installation

1. Install the required dependencies:

pip install mcp

Usage

Step 1: Start the Feedback Server

Run the server in a terminal:

python feedback_server.py

The server will:

• Start on localhost:9876

• Display a message when ready

• Show agent requests and allow you to provide feedback interactively

Step 2: Configure Your MCP Agent

Add the server to your MCP configuration file (e.g., mcp_config.json):

{
  "mcpServers": {
    "feedback-server": {
      "command": "python",
      "args": [
        "/<path_to_script>/feedback_client.py"
      ]
    }
  }
}

Step 3: Agent Prompt

Use this prompt with your agent:

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

How It Works

1. Agent calls the tool: When an agent needs feedback, it calls the interactive_feedback tool with:

   • work_summary: Summary of work completed so far

   • question (optional): Specific question for the user

2. Server displays request: The feedback server shows:

   • Timestamp of the request

   • Work summary from the agent

   • Any specific questions

3. User provides feedback: In the terminal running the server:

   • Type feedback and press Enter to send it back to the agent

   • Press Enter with empty input to approve and let the agent continue

4. Agent receives response: The agent gets either:

   • User feedback to act upon

   • Approval to continue (when feedback is empty)

Example Interaction

In the server terminal:

🚀 Feedback Server started on localhost:9876
Waiting for agent connections...

==============================================================
📝 AGENT REQUEST - 2025-01-15 14:30:45
==============================================================

Work Summary:
I have completed the following tasks:
1. Created the user authentication system
2. Set up the database models
3. Implemented the API endpoints

Agent's Question:
Should I proceed with adding the frontend components?

==============================================================
📌 Your Feedback (press Enter with empty input to approve and continue):
> Yes, but make sure to use React with TypeScript
✅ Feedback sent to agent: 'Yes, but make sure to use React with TypeScript'

Features

• ✅ Real-time interactive feedback

• ✅ Socket-based communication (no polling)

• ✅ Clear visual feedback in terminal

• ✅ Support for both general feedback and specific questions

• ✅ Simple approval mechanism (empty input = continue)

• ✅ Error handling and connection management

Troubleshooting

• Connection refused: Make sure the feedback server is running before the agent tries to connect

• Port already in use: The server uses port 9876 by default. Make sure no other process is using this port

• MCP not found: Install the MCP package using pip install mcp

Architecture

The system uses a dual-server architecture:

1. MCP Server: Handles the MCP protocol and tool definitions

2. Socket Server: Manages the interactive feedback loop in the terminal

This design allows for real-time interaction while maintaining compatibility with the MCP protocol.