MCP Sandbox Project

Description

This project is a sandbox environment showcasing various tools and functionalities implemented using the Model Context Protocol (MCP) and the FastMCP library. It includes several independent MCP servers, each providing specific capabilities.

Features / Components

This project includes the following MCP servers:

• tool_greeting.py: Provides simple tools like echo and a greeting tool that uses a dynamic resource.
• tool_disk.py: Offers tools for basic disk operations like creating/deleting folders and reading/writing/deleting files within a base directory (c:/workspace/test).
• tool_excel.py: Includes a tool to create Excel files (.xlsx) with specified content in the base directory (c:/workspace/test).
• tool_keyboard.py: Provides a tool to control the keyboard, specifically to type a message into Notepad.
• tool_hwp.py: Contains a tool to read text content from Hangul Word Processor (.hwp) files located in the base directory (c:/workspace/test).
• tool_math.py: A simple server with an add tool and a dynamic goodbye resource.
• prompts.py: Defines custom prompts (prompt_decision, prompt_opinion) that can be used with language models.
• resource_agent.py: Implements a resource (Agent://state) to retrieve the state of an AI agent.
• main.py: A simple placeholder script.

Setup

1. Prerequisites: Ensure you have Python (>=3.12) and uv installed.
2. Dependencies: Install the required Python packages listed in pyproject.toml. You can typically do this using uv:

   uv pip install -r requirements.txt 
   

   (You might need to generate requirements.txt from pyproject.toml first using uv pip freeze > requirements.txt or install directly using uv pip install . if your setup supports it).

Usage

Each MCP server can be run individually. The configuration in ~\.cursor\mcp.json suggests these servers are intended to be launched via uv. See save/claude_desktop_config.json for reference.

To run a specific server manually (example):

uv run tool_greeting.py
mcp dev tool_greeting.py

Replace tool_greeting.py with the script name of the server you want to run. These servers can then be accessed by MCP clients or frameworks that integrate with MCP.

Configuration

• Base Directories: Some tools (tool_disk.py, tool_excel.py, tool_hwp.py) operate within a specific base directory (c:/workspace/test). Ensure this directory exists or modify the BASE_DIR constant in the respective scripts if needed.
• Server Ports: Each server is configured to run on a specific port (e.g., tool_greeting on 5000, tool_disk on 5001, etc.). Ensure these ports are available.

References

MCP SDK

• https://github.com/modelcontextprotocol
• https://github.com/modelcontextprotocol/swift-sdk
• https://pypi.org/project/mcp/ (python)

MCP server repos

• https://github.com/modelcontextprotocol/servers/tree/main
• https://mcp.so
• https://smithery.ai
• https://www.pulsemcp.com
• https://www.npmjs.com/search?q=mcp
• https://cursor.directory/mcp

MCP framework or extentions.

• https://fast-agent.ai (fast-agent - MCP native Agents and Workflows)
• https://github.com/langchain-ai/langchain-mcp-adapters (Langchain)
• https://github.com/tadata-org/fastapi_mcp (FastAPI)
• https://www.youtube.com/watch?v=eczLRG3q4V0 (Flutter)
• (cf) https://github.com/settings/copilot

Use cases of 3rd parties

• https://gitmcp.io (Instantly create a Remote MCP server for any GitHub repository)
• https://github.com/elevenlabs/elevenlabs-mcp/tree/main (elevenlabs)
• https://docs.windsurf.com/windsurf/mcp
• https://www.youtube.com/watch?v=dutyOc_cAEU (VS code)
• https://gist.github.com/burkeholland/24802296b5bfaaf7fb775c81cd626512

Other references

• https://modelcontextprotocol.io/quickstart/server
• https://www.books.weniv.co.kr/basecamp-mcp
• https://modelcontextprotocol.io/clients (Feature support matrix)
• https://www.youtube.com/watch?v=EswVjHZMn74

Roadmap:

• https://modelcontextprotocol.io/development/roadmap
• OAuth Authentication (not API keys)
• MCP registry