my-mcp
An MCP server framework featuring dynamic tool loading and a modular one-tool-per-file architecture for rapid development. It supports both Stdio and HTTP transport modes, offering automated test generation and centralized configuration management.
README
my-mcp
my-mcp is a Model Context Protocol (MCP) server built with FastMCP featuring dynamic tool loading.
Features
- Dynamic Tool Loading: Tools are automatically discovered and loaded from
src/tools/ - One Tool Per File: Each tool is a single file with a function matching the filename
- FastMCP Integration: Leverages FastMCP for robust MCP protocol handling
- Configuration Management: Tool-specific configuration via
mcp.yaml - Fail-Fast: Server won't start if any tool fails to load
- Auto-Generated Tests: Automatic test generation for tool validation
Project Structure
src/
├── tools/ # Tool implementations (one file per tool)
│ ├── echo.py # Example echo tool
│ └── __init__.py # Auto-generated tool registry
├── core/ # Dynamic loading framework
│ ├── server.py # Dynamic MCP server
│ └── utils.py # Shared utilities
└── main.py # Entry point
mcp.yaml # Configuration file
tests/ # Generated tests
Quick Start
Option 1: Local Development (with Python/uv)
-
Install Dependencies:
uv sync -
Run the Server:
# Stdio mode (default MCP transport) uv run python src/main.py # HTTP mode with WebSocket MCP endpoint uv run python src/main.py --http # HTTP mode with custom host/port uv run python src/main.py --http --host 0.0.0.0 --port 8080 -
Using uv Scripts:
# Development mode (HTTP on port 3000) uv run dev # HTTP mode uv run dev-http # Stdio mode uv run start -
Add New Tools:
# Create a new tool (no tool types needed!) arctl mcp add-tool weather # The tool file will be created at src/tools/weather.py # Edit it to implement your tool logic
Option 2: Docker-Only Development (no local Python/uv required)
-
Build Docker Image:
arctl mcp build --verbose -
Run in Container:
docker run -i my-mcp:latest -
Add New Tools:
# Create a new tool arctl mcp add-tool weather # Edit the tool file, then rebuild arctl mcp build
HTTP Transport Mode
The server supports running in HTTP mode for development and integration purposes.
Starting in HTTP Mode
# Command line flag
python src/main.py --http
# Environment variable
MCP_TRANSPORT_MODE=http python src/main.py
# Custom host and port
python src/main.py --http --host localhost --port 8080
Creating Tools
Basic Tool Structure
Each tool is a Python file in src/tools/ containing a function decorated with @mcp.tool():
# src/tools/weather.py
from core.server import mcp
from core.utils import get_tool_config, get_env_var
@mcp.tool()
def weather(location: str) -> str:
"""Get weather information for a location."""
# Get tool configuration
config = get_tool_config("weather")
api_key = get_env_var(config.get("api_key_env", "WEATHER_API_KEY"))
base_url = config.get("base_url", "https://api.openweathermap.org/data/2.5")
# TODO: Implement weather API call
return f"Weather for {location}: Sunny, 72°F"
Tool Examples
The generated tool template includes commented examples for common patterns:
# HTTP API calls
# async with httpx.AsyncClient() as client:
# response = await client.get(f"{base_url}/weather?q={location}&appid={api_key}")
# return response.json()
# Database operations
# async with asyncpg.connect(connection_string) as conn:
# result = await conn.fetchrow("SELECT * FROM weather WHERE location = $1", location)
# return dict(result)
# File processing
# with open(file_path, 'r') as f:
# content = f.read()
# return {"content": content, "size": len(content)}
Configuration
Configure tools in mcp.yaml:
tools:
weather:
api_key_env: "WEATHER_API_KEY"
base_url: "https://api.openweathermap.org/data/2.5"
timeout: 30
database:
connection_string_env: "DATABASE_URL"
max_connections: 10
Testing
Run the generated tests to verify your tools load correctly:
uv run pytest tests/
Development
Adding Dependencies
Update pyproject.toml and run:
uv sync
Code Quality
uv run black .
uv run ruff check .
uv run mypy .
Deployment
Docker
# Build image (handles lockfile automatically)
arctl mcp build
# Run container
docker run -i my-mcp:latest
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。