mcpserver.so LogoMCPFly
Submit
MCP Server logo

MCP Server

by tikoehle

AI/MLServerClientSTDIOSSEPython

A simple MCP (Model Context Protocol) server implementation with examples for STDIO and SSE transports. It includes instructions for setting up the server, client, and inspector, along with testing and usage examples.

View on GitHub

Last updated: N/A

Simple MCP Client / Server

https://modelcontextprotocol.io/quickstart/server

Install uv, on macOS and Linux.

$ curl -LsSf https://astral.sh/uv/install.sh | sh

Then git clone the repo.

MCP Server (stdio)

uv init mcp_server
cd mcp_server/
uv run main.py                        # just to create the .venv, then rm the main.py

uv add "mcp[cli]" httpx               # MCP server Python lib

touch weather_stdio.py                # STDIO server implementation

Start the server

mcp_server$ uv run weather_stdio.py

Test with VSCode + Cline

Add MCP server to Cline.

Cline -> MCP Servers (icon close to '+') -> Installed -> Add

You get a json template "mcpServers": {} and you add the new server. Here is the cline_mcp_settings.json:

{
  "mcpServers": {
    "weather": {
            "command": "uv",
            "args": [
                "--directory",
                "/home/users/tikoehle/mcp_server",
                "run",
                "weather_stdio.py"
            ]
        }
  }
}

=> Cline shows the 2 get_* Tools functions. Click 'Done' to return to Cline Task input trying the new server.

Test

Type a task calling the two tools: "Can you tell me the weather forecast for San Francisco, CA" "Can you tell me the weather alerts for San Francisco, CA"

=> uses my two new MCP 'weather' server tools.

Note: Local MCP servers should not log messages to stdout because this will interfere with protocol operation.

MCP Client

https://modelcontextprotocol.io/quickstart/client

uv init mcp_client
cd mcp_client/
uv run main.py                     # create venv

uv add mcp

touch client_stdio.py              # MCP client impl.

Run the client

uv run client.py path/to/server.py

mcp_client$ uv run client_stdio.py ../mcp_server/weather_stdio.py

MCP Inspector

Installation Node.js

On the MCP server compute, get Node.js (Current) for Linux using nvm with npm. https://nodejs.org/en/download

cd $HOME

# Download and install nvm:
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.2/install.sh | bash

# in lieu of restarting the shell
\. "$HOME/.nvm/nvm.sh"

# Download and install Node.js:
nvm install 23

# Verify the Node.js version:
node -v # Should print "v23.10.0".
nvm current # Should print "v23.10.0".

# Verify npm version:
npm -v # Should print "10.9.2".

Inspecting a locally developed server

mcp_server$ uv run mcp dev ./weather_stdio.py
Starting MCP inspector...
Proxy server listening on port 3000

šŸ” MCP Inspector is up and running at http://localhost:5173 šŸš€

Note:

Another method to install the latest version and start the Inspector.

mcp_server$ npx @modelcontextprotocol/inspector@latest

Open the MCP Inspector client UI in the Browser

http://comp9:5173   --> Connect

Note:

If the MCP dev server runs on a remote compute, for example comp9, then the MCP Inspector Client UI needs to connect to this machine.

Transport Type, Command and Arguments appeared with the correct parameters when launching the UI in the browser.

Transport Type: STDIO
Command: uv
Arguments: run --with mcp mcp run ./weather_stdio.py

or

Transport Type: SSE
URL: http://comp9:8001/sse

Note:

MCP development tools

mcp_server$ uv run mcp

SSE Client / Server

Server

mcp_server$ uv run weather_sse.py      # port 8001

Client

mcp_client$ uv run client_sse.py http://comp9:8001/sse

Python-SDK

https://github.com/modelcontextprotocol/python-sdk/

SSE Issues:

https://github.com/modelcontextprotocol/python-sdk/issues/398