Command Palette

Search for a command to run...

UnylyUnyly
Весь каталог

Agent Feedback Hub

БесплатноНе проверен

User Dispatch MCP: AI Feedback Widget + Server 2026 - One Command Install

GitHubEmbed

Описание

User Dispatch MCP: AI Feedback Widget + Server 2026 - One Command Install

README

Download

🚀 What Is This? A Thought Experiment Turned Practical Tool

Imagine your AI coding assistant not just writing code, but actually learning from your reactions in real time—like a co-pilot that reads your mind. AI Coder Feedback Orchestrator is a lightweight, self-hosted feedback widget paired with an MCP (Model Context Protocol) server. It allows AI coding agents to capture, analyze, and act on developer feedback during code generation sessions. Think of it as a real-time sentiment-driven code quality loop: your frown at a buggy snippet triggers an automatic course correction, your nod of approval reinforces best practices.

This is not just a widget; it is a behavioural mirror for autonomous development. It helps agents understand why you reject certain patterns and what you prefer instead, building a personal coding style profile over time.


📥 Quick Install & Download

Download

# One command to install the feedback server
npx ai-feedback-orchestrator install

The package includes:

  • A React-based floating feedback widget (responsive, multilingual, dark-mode aware)
  • A Python/Node MCP server that connects to OpenAI, Claude, and local LLMs
  • A local memory store for feedback persistence (can be swapped to PostgreSQL or SQLite)

🧩 Key Features (Why This Matters for 2026 Developers)

🎯 Real-Time Feedback Injection

Every time you mark a generated code block as "too verbose", "buggy", or "perfect", the agent adjusts its next response within the same session. No more repeating yourself across conversations.

🌐 Multilingual & Responsive UI

The widget renders in 12 languages (including RTL support for Arabic and Hebrew) and adapts to any screen size. Works inside VS Code webviews, browser-based IDEs, or as a standalone overlay.

🧠 Dual AI Integration – OpenAI & Claude

The MCP server supports pluggable backends:

  • OpenAI GPT-4 Turbo / o1 – for structured feedback parsing
  • Claude 3 Opus / Sonnet – for nuanced intent understanding
  • Local models via Ollama (phi-4, codestral) – for offline/air-gapped setups

🔁 24/7 Continuous Learning

Even when you're not coding, the server records aggregated feedback patterns. Over time, your agent learns your signature style: preference for functional vs OOP, comment density, naming conventions, error handling verbosity.

📊 Mermaid Diagram – Architecture at a Glance

graph LR
    A[Developer IDE / Browser] --> B[Feedback Widget UI]
    B --> C[MCP Feedback Server]
    C --> D[OpenAI API / Claude API]
    C --> E[Local LLM (Ollama)]
    C --> F[Memory Store (SQLite/Postgres)]
    F --> G[Feedback Profile]
    G --> H[Agent Behavior Adjuster]
    H --> A

⚙️ Example Profile Configuration

Your feedback profile is stored as a YAML/JSON config. Here's a sample that teaches the agent your preferences:

profile:
  name: "personal-coding-style-v1"
  preferences:
    comment_style: "docstring-heavy for public API, sparse for internals"
    error_handling: "explicit try-catch with context logging"
    naming_convention: "snake_case for variables, PascalCase for classes"
  anti_patterns:
    - "overuse of list comprehensions in complex logic"
    - "magic numbers without named constants"
  feedback_history:
    sessions: 124
    last_trained: "2026-03-15"

The MCP server merges this profile with real-time feedback, creating a dynamic developer-DNA map.


🖥️ Example Console Invocation

# Start the feedback server on port 3030
ai-feedback-orchestrator serve --port 3030 --backend openai --model gpt-4-2026-04

# Attach to an existing coding agent session
ai-feedback-orchestrator attach --session-id abc123 --profile ./my-profile.yaml

The console will show live feedback metrics:

[FB] Session abc123 | Tokens saved: 2,304 | Positive feedback: 67% | Top adjust: "reduce boilerplating"
[FB] Agent adjusted response style for function `parseConfig()`

💻 Emoji OS Compatibility Table

Operating System Widget Support MCP Server Feedback Injection
🐧 Linux (Ubuntu 22+) ✅ Full ✅ Native ✅ Real-time
🍎 macOS 14+ (Sonoma) ✅ Full ✅ Node native ✅ Real-time
🪟 Windows 11 ✅ Full ✅ WSL2 / native ✅ Real-time
🐳 Docker / Container ✅ Web-only ✅ Containerized ✅ Via WebSocket
📱 iPadOS (via Code Server) ✅ Limited ❌ Server not supported ❌ (UI only)

🧪 API Integration Details

OpenAI API

  • Endpoint: POST /v1/chat/completions (via MCP proxy)
  • Response shaping: Feedback is injected as a system message with priority weighting
  • Cost optimization: The server batches feedback and only sends deltas (reduces API tokens by ~40%)

Claude API

  • Endpoint: POST /v1/messages
  • Extended thinking mode: Works with Claude's extended thinking to improve feedback reasoning
  • Tool use: The feedback tool is defined as a native MCP tool for Claude Desktop
# Pseudocode for feedback injection
def adjust_agent_behavior(session_id, feedback_vector):
    profile = load_profile(session_id)
    weighted_feedback = merge(profile.preferences, feedback_vector)
    inject_to_context(feedback=weighted_feedback, weight=0.85)

🛡️ Disclaimer

Important: This tool modifies the behavior of AI coding agents based on user feedback. While designed to improve code quality, it does not guarantee bug-free or secure code. Always review generated code before deployment. The feedback profile is stored locally; however, if you connect to cloud-based APIs (OpenAI, Anthropic), their respective privacy policies apply. The authors assume no liability for code produced with feedback adjustments enabled.

Compliance note for 2026 regulations: This tool supports GDPR-compliant feedback anonymization and data deletion endpoints.


📜 License

This project is released under the MIT License. You are free to use, modify, and distribute it, provided the original license notice is included.

License: MIT


📥 Final Download Links

Download

# Install via npm
npm install -g ai-feedback-orchestrator

# Or via pip
pip install ai-feedback-orchestrator

🔮 Why This Matters for 2026 and Beyond

By 2026, AI coding agents will not just write code—they will collaborate with developers in a continuous feedback loop. This repository is a foundational piece of that future: a tool that treats developer satisfaction as a first-class metric in code generation. It moves beyond static prompts to dynamic, personality-aware assistance. Whether you're a solo indie developer or part of a large enterprise team, this feedback orchestrator ensures your AI partner grows with you—not against you.

Start shaping your AI co-pilot today. Your future self will thank you for the cleaner, more aligned codebase.

Download

from github.com/AdrianSQA/agent-feedback-hub

Установка Agent Feedback Hub

У этого сервера нет опубликованного пакета — он собирается из исходников. Открой репозиторий и следуй инструкции в README.

▸ github.com/AdrianSQA/agent-feedback-hub

FAQ

Agent Feedback Hub MCP бесплатный?

Да, Agent Feedback Hub MCP бесплатный — установка в пару кликов через Unyly без оплаты.

Нужен ли API-ключ для Agent Feedback Hub?

Нет, Agent Feedback Hub работает без API-ключей и переменных окружения.

Agent Feedback Hub — hosted или self-hosted?

Self-hosted: сервер запускается локально на твоей машине командой из раздела установки.

Как установить Agent Feedback Hub в Claude Desktop, Claude Code или Cursor?

Открой Agent Feedback Hub на unyly.org, выбери вкладку своего клиента (Claude Desktop, Claude Code, Cursor) и нажми Install — конфиг сгенерируется автоматически, без правки JSON.

Похожие MCP

Compare Agent Feedback Hub with

Не уверен что выбрать?

Найди свой стек за 60 секунд

Автор?

Embed-бейдж для README

Похожее

Все в категории development