README.md

---
title: The True C-Codestral IDE Agent
emoji: 🚀
colorFrom: red
colorTo: red
sdk: gradio
sdk_version: 5.33.1
app_file: app.py
pinned: true
license: mit
tags:
- agent-demo-track
- mcp-server-track
---

<div align="center">

# 🚀 The True C-Codestral IDE Agent

**An intelligent, multi-step AI agent that transforms natural language instructions into compiled C code, acting as a complete, interactive C development environment.**

[![Hackathon Track](https://img.shields.io/badge/Hackathon%20Track-Agentic%20Demo-blue)](https://huggingface.co/spaces/Agents-MCP-Hackathon/c)
[![Hackathon Track](https://img.shields.io/badge/Hackathon%20Track-MCP%20Server-orange)](https://huggingface.co/spaces/Agents-MCP-Hackathon/c)

</div>

This project is a submission for the **Agents & MCP Hackathon 2025**, participating in both the **Agentic Demo Showcase** and the **MCP Tool / Server** tracks.

---

## 🎥 Demo Video
[![C-Codestral Agent Demo Video](https://img.youtube.com/vi/LuO--JG7t7o/0.jpg)](https://www.youtube.com/watch?v=LuO--JG7t7o)
**(Click the image to watch the demo on YouTube)**

---

## ✨ Core Features / 핵심 기능

This agent is more than a simple code generator; it's a true AI development partner that understands and executes complex, sequential tasks.

*   **🧠 Multi-Step Agentic Workflow**
    *   **EN**: Understands and sequentially executes chained commands like "Refactor this code for readability, and then compile it."
    *   **KR**: "코드를 리팩토링하고, 그 다음에 컴파일해줘"와 같은 복잡하고 연결된 명령을 이해하고 순차적으로 작업을 수행합니다.

*   **🤖 Powered by Codestral**
    *   **EN**: Utilizes Mistral AI's latest specialized model for code, `Codestral`, to generate and analyze high-quality C code.
    *   **KR**: Mistral AI의 최신 코드 전문 모델인 `Codestral`을 사용하여 정확하고 품질 높은 C 코드를 생성하고 분석합니다.

*   **⚡ On-the-Fly Compilation & Execution**
    *   **EN**: Instantly compiles and runs code using `gcc` on the backend. It provides real-time feedback on the console, including success status, output, or compilation errors.
    *   **KR**: 백엔드에서 `gcc`를 통해 코드를 즉시 컴파일하고 실행합니다. 성공 여부, 결과, 컴파일 오류까지 실시간으로 콘솔에 표시하여 완벽한 C 개발 경험을 제공합니다.

*   **↔️ Dual-Mode Interface**
    *   **EN**: Features both a conversational **IDE Agent** for users and **MCP Tools API** endpoints for other agents.
    *   **KR**: 사용자를 위한 대화형 **IDE Agent**와 다른 에이전트들이 사용할 수 있는 **MCP Tools API**를 모두 제공합니다.

*   **🌍 Bilingual Instruction Support**
    *   **EN**: Capable of understanding and processing instructions given in both English and Korean.
    *   **KR**: 한국어와 영어로 작성된 명령어를 모두 이해하고 처리할 수 있습니다.

---

## 🎯 Hackathon Tracks Contribution / 해커톤 트랙 기여

This project makes clear contributions to two separate tracks.

### 🤖 Track 3: Agentic Demo Showcase

The **IDE Agent** tab is a complete demonstration of a powerful AI agent.
1.  **Stateful Interaction**: The agent maintains the code (state) throughout the conversation, building upon previous steps.
2.  **Complex Command Parsing**: It deconstructs user requests based on connectors like "and" or "then" to create a multi-step execution plan.
3.  **Autonomous Tool Use**: Based on its plan, the agent autonomously selects and uses its internal tools: `generate_code`, `compile_and_run`, and `analyze_and_refactor`.

These capabilities prove it is a true "agent," moving beyond the limitations of a simple chatbot.

### 🔧 Track 1: MCP Tool / Server

The **MCP Tools API** tab demonstrates how a Gradio app can function as an MCP (Model Context Protocol) server.
*   Each function in this tab (`Generate C Code`, `Compile & Run C Code`, etc.) acts as a discrete API endpoint that can be called by any MCP client or other agents.
*   This exposes the agent's powerful C language processing capabilities as modular tools, allowing other AI services to easily integrate them.

---

## 🛠️ How It Works & Technical Details / 기술 스택 및 작동 방식

*   **Frontend**: Built with `Gradio` to create a modern, interactive web UI. The interface is cleanly separated into two tabs for its distinct purposes (IDE and MCP).
*   **LLM Backend**: Leverages `Mistral AI`'s `Codestral-latest` model via API calls to handle the core intelligence of code generation, analysis, and refactoring.
*   **Execution Backend**: Uses Python's `subprocess` module to run the `gcc` compiler in a secure environment, allowing it to feed actual C code execution results back to the user.
*   **Agent Logic**: The agent's core logic parses user commands with regular expressions to create a task queue. It then iterates through this queue, executing each task sequentially. Results and logs from each step are streamed to the user in real-time for full transparency.

## 🚀 How to Use / 사용 방법

1.  **IDE Agent Tab**:
    *   **EN**: Write your C code in the **C Code Editor** on the left, or leave it empty to ask the agent to generate new code. In the **Instruction** box on the right, type your command in natural language (English or Korean), e.g., "Create a program that prints 'Hello World' and then run it." Click **Execute**. The agent will perform the tasks, stream its progress to the **Console / Output**, and update the code editor automatically if needed.
    *   **KR**: 왼쪽의 **C Code Editor**에 코드를 작성하거나, 비워두고 새로운 코드 생성을 요청할 수 있습니다. 오른쪽 **Instruction** 박스에 원하는 작업을 자연어(한국어 또는 영어)로 입력합니다. (예: "Hello World를 출력하는 코드를 만들고 실행해줘"). **Execute** 버튼을 누르면 에이전트가 작업을 수행하며 **Console / Output**에 진행 과정과 최종 결과를 표시하고, 필요시 코드 편집기 내용을 자동으로 업데이트합니다.

2.  **MCP Tools API Tab**:
    *   **EN**: This tab is designed for other agents or services to use this Space as an API. You can open each tool's accordion to test its functionality individually.
    *   **KR**: 이 탭은 다른 에이전트나 서비스가 이 Space를 API처럼 사용할 때를 위한 것입니다. 각 도구(Tool) 섹션을 열어 개별 기능을 테스트해볼 수 있습니다.