MCP-123
The easiest way to run an MCP server & client (2 lines each)
🚀 Features
-
Ultra-minimal setup: Start a server or client in 2 lines.
-
Easy tool creation: Write normal functions in your
tools.pyfile—no decorators or special wrappers needed—and they get included as tools that your MCP server can use automatically. -
OpenAI integration: The client uses your OpenAI API key to answer questions, calling tools as needed.
🖥️ Quickstart
1. Install Requirements
pip install -r requirements.txt
2. Create Your Tools
Define your functions in tools.py. No decorators needed, they are automatically added to your MCP server as tools. For example:
def add(a: int, b: int) -> int:
"""Add two numbers."""
return a + b
3. Start the MCP Server (2 lines)
from mcp123 import server
server.run_server("tools.py", port=9999)
4. Set up the MCP Client (2 lines)
from mcp123.client import McpClient
client = McpClient("http://localhost:9999", "sk-...your OpenAI key...")
5. Use the MCP Client
answer = client.ask("Add 15 and 14.")
print("Answer:", answer)
6. Close the MCP Client when you are done
client.close()
🚀 Features
-
Ultra-minimal setup: Start a server or client in 2 lines.
-
Easy tool creation: Write normal functions in your
tools.pyfile—no decorators or special wrappers needed—and they get included as tools that your MCP server can use automatically. -
OpenAI integration: The client uses your OpenAI API key to answer questions, calling tools as needed.
📝 How It Works
-
Server: Loads all top-level functions from
tools.pyand exposes them as MCP tools via HTTP. -
Client: Discovers available tools, sends prompts to OpenAI, and automatically calls tools if needed.
🛠️ Example Output
When you run the client, you’ll see:
Tools discovered:
[ ...list of tools... ]
Answer: 29
🔑 Requirements
-
Python 3.11+
-
OpenAI API key (for the client)
📢 Why MCP123?
-
Zero boilerplate: No need to write schemas or wrappers—just write functions.
-
LLM-native: Designed for seamless LLM tool use.
-
Extensible: Add more tools by simply adding functions.
🤝 Credits
-
Built with FastMCP
-
Inspired by the Model Context Protocol (MCP)
📬 Feedback & Contributions
Pull requests and issues are welcome, but only if they are in ALL-CAPS.