---

Table of Contents

• Overview
• Quick Start
• Tech Stack
• Architecture
• Features
• Prerequisites
• Installation
• Configuration
• Usage
  • stdio MCP Server
  • SSE MCP Server
  • AI Agent Chat
  • Docker
• MCP Tools Reference
• Project Structure
• Troubleshooting
• Contributing
• License

---

Overview

This project is a complete, runnable example of an MCP server built with Python and FastMCP. It surfaces live US weather data from the NOAA National Weather Service API as callable MCP tools, allowing any MCP-compatible client — Claude Desktop, VS Code extensions, or a custom agent — to discover and invoke them.

Two implementations run side-by-side so you can compare approaches:

Directory	Transport	Client side
server/	stdio	AI agent via mcp_use + Groq LLM
mcp-server/	SSE (HTTP) + stdio	Raw ClientSession

---

Quick Start

# 1. Clone and install
git clone https://github.com/<your-username>/mcp-lab.git
cd mcp-lab
uv sync

# 2. Add your Groq key
echo "GROQ_API_KEY=your_key_here" > .env

# 3. Run the AI agent chat
uv run server/client.py

You: Are there any weather alerts in Texas?
Assistant: Yes — there is currently a Heat Advisory in effect for...

You: What's the forecast for New York City?
Assistant: Here is the 5-period forecast for NYC (40.71, -74.00)...

You: clear    # clear conversation history
You: exit     # quit

No weather API key required — NOAA NWS is a free, public API.

---

Tech Stack

Layer	Tool
MCP framework	FastMCP via mcp[cli]
AI agent orchestration	mcp_use
LLM	Qwen QwQ-32B via Groq (langchain-groq)
HTTP client	httpx
Weather data	NOAA National Weather Service REST API
Dependency management	uv
Containerisation	Docker

---

Architecture

┌──────────────────────────────────────────────────────────────┐
│                        MCP Clients                           │
│                                                              │
│  ┌─────────────────────┐      ┌────────────────────────────┐ │
│  │   server/client.py  │      │  mcp-server/client-sse.py  │ │
│  │   ───────────────── │      │  mcp-server/client-stdio.py│ │
│  │   MCPAgent          │      │  ────────────────────────  │ │
│  │   Groq (Qwen) LLM   │      │  Raw MCP ClientSession     │ │
│  └──────────┬──────────┘      └────────────┬───────────────┘ │
└─────────────┼──────────────────────────────┼──────────────── ┘
              │ stdio (subprocess)           │ SSE (HTTP) / stdio
              ▼                              ▼
┌─────────────────────────┐   ┌──────────────────────────────┐
│  server/weather.py      │   │  mcp-server/server.py        │
│  ────────────────────── │   │  ──────────────────────────  │
│  FastMCP (stdio)        │   │  FastMCP (SSE, port 8000)    │
│  Tools: get_weather_    │   │  Tools: get_alerts           │
│         alerts          │   │         get_forecast         │
└──────────┬──────────────┘   └──────────────┬───────────────┘
           │                                 │
           └──────────────┬──────────────────┘
                          ▼
              ┌───────────────────────┐
              │     NOAA NWS REST API │
              │  api.weather.gov      │
              └───────────────────────┘

Data flow (AI agent path):

User prompt
    → MCPAgent (mcp_use)
        → Groq LLM decides which MCP tool to call
            → FastMCP server executes tool
                → NOAA NWS API (HTTP)
            ← Raw weather JSON
        ← Formatted tool result
    ← Natural-language response
User

---

Features

• Weather alerts — fetch active NOAA alerts for any US state by two-letter code
• Weather forecast — get a multi-period forecast for any US lat/lon coordinate
• Two transport modes — stdio for subprocess embedding, SSE for HTTP-based remote access
• AI agent chat loop — interactive conversation; the LLM automatically decides when to call tools
• Persistent conversation memory — multi-turn context retained across the session
• Docker support — containerise the SSE server in one command
• No weather API key — uses the free public NOAA NWS API

---

Prerequisites

Requirement	Version	Notes
Python	3.13+	Required by pyproject.toml
uv	latest	Recommended package manager
Groq API key	—	Free tier available; needed only for AI agent chat
Docker	any recent	Optional; only for containerised SSE server

---

Installation

Using uv (recommended)

git clone https://github.com/<your-username>/mcp-lab.git
cd mcp-lab
uv sync

Using pip

git clone https://github.com/<your-username>/mcp-lab.git
cd mcp-lab
pip install -r requirements.txt

---

Configuration

Copy the example env file and fill in your Groq key:

cp .env.example .env   # or create .env manually

.env:

GROQ_API_KEY=your_groq_api_key_here

Note: The NOAA weather API is public and requires no credentials. GROQ_API_KEY is only needed for the AI agent chat (server/client.py).

---

Usage

1. stdio MCP Server

server/weather.py communicates over standard I/O. It is designed to be spawned by an MCP host (e.g. Claude Desktop) or programmatically by the AI agent.

uv run server/weather.py

To register it with an MCP host, inspect the bundled config:

cat server/weather.json   # update the absolute path if needed

---

2. SSE MCP Server (HTTP)

mcp-server/server.py exposes the same tools over HTTP using Server-Sent Events on port 8000.

# Terminal 1 — start the server
uv run mcp-server/server.py

# Terminal 2 — connect with the SSE client
uv run mcp-server/client-sse.py

# Or use the stdio client (spawns server as a subprocess)
uv run mcp-server/client-stdio.py

The server is available at http://localhost:8000.

---

3. AI Agent Chat

An interactive chat loop that understands plain-English weather queries. The Groq-hosted Qwen model automatically decides when to call the weather tools.

uv run server/client.py

Example session:

===== Interactive MCP Chat =====
Type 'exit' or 'quit' to end the conversation
Type 'clear' to clear conversation history
==================================

You: Are there any weather alerts in California?
Assistant: Yes, there is currently a Fire Weather Watch in effect for...

You: What is the forecast for Chicago?
Assistant: Here is the forecast for Chicago (41.88, -87.63)...

You: clear
Conversation history cleared.

You: exit
Ending conversation...

---

4. Docker

Build and run the SSE server as a container:

cd mcp-server

# Build the image
docker build -t mcp-weather-server .

# Run (SSE server accessible at http://localhost:8000)
docker run -p 8000:8000 mcp-weather-server

---

MCP Tools Reference

Both server implementations expose the same logical tools under slightly different names.

---

get_weather_alerts · get_alerts

Fetch active NOAA weather alerts for a US state.

Parameter	Type	Required	Description
state	str	Yes	Two-letter US state code — e.g. CA, TX, NY

Returns: A formatted string listing each active alert with event type, affected area, severity, description, and instructions. Returns "No active alerts for this state." when none exist.

Example:

get_alerts("TX")
# Event: Heat Advisory
# Area: South Central Texas
# Severity: Moderate
# ...

---

get_forecast

Fetch the next five forecast periods for a US latitude/longitude coordinate.

Parameter	Type	Required	Description
latitude	float	Yes	Latitude of the target location
longitude	float	Yes	Longitude of the target location

Returns: A formatted string with up to 5 forecast periods, each including period name, temperature, wind speed/direction, and a detailed forecast description.

Coverage: NOAA NWS only covers US locations. Coordinates outside the US will return an error.

Example:

get_forecast(40.71, -74.00)
# Tonight:
# Temperature: 68°F
# Wind: 10 mph S
# Forecast: Partly cloudy, with a low around 68...

---

Project Structure

mcp-lab/
│
├── server/                     # stdio transport — AI agent demo
│   ├── weather.py              # FastMCP server (stdio)
│   ├── client.py               # AI agent chat (mcp_use + Groq LLM)
│   └── weather.json            # MCP host config (Claude Desktop, etc.)
│
├── mcp-server/                 # SSE/HTTP transport — raw client demo
│   ├── server.py               # FastMCP server (SSE, port 8000)
│   ├── client-sse.py           # Raw MCP client over SSE
│   ├── client-stdio.py         # Raw MCP client over stdio
│   └── Dockerfile              # Container image for the SSE server
│
├── main.py                     # Entry point placeholder
├── pyproject.toml              # Project metadata and dependencies (uv/pip)
├── requirements.txt            # pip-compatible dependency list
├── uv.lock                     # Locked dependency graph (uv)
├── .env.example                # Environment variable template
├── .gitignore
└── README.md

---

Troubleshooting

GROQ_API_KEY not found

KeyError: 'GROQ_API_KEY'

Make sure a .env file exists in the project root with a valid key. The AI agent chat loads it automatically via python-dotenv.

---

NOAA API returns no data for a location

The NOAA NWS API only covers US locations. Coordinates outside the US (or offshore) will return an error from the /points endpoint. Verify your lat/lon is within the continental US, Alaska, or Hawaii.

---

SSE client can't connect to localhost:8000

Make sure mcp-server/server.py is running in a separate terminal before launching client-sse.py. Check the server logs for any port-binding errors.

---

uv sync fails on Python version

This project requires Python 3.13+. Check your active version:

python --version
uv python list

Install a compatible interpreter with uv python install 3.13.

---

Docker build fails at uv pip install

The Dockerfile uses Python 3.11-slim as its base image to keep the image lean; this is separate from the local dev requirement of 3.13+. If you see dependency resolution errors, confirm your requirements.txt is up to date:

uv pip compile pyproject.toml -o requirements.txt

---

Contributing

Contributions, issues, and feature requests are welcome.

1. Fork the repository
2. Create a branch: git checkout -b feature/my-feature
3. Make your changes — follow PEP 8 style
4. Test your changes locally before opening a PR
5. Commit: git commit -m "feat: add my feature"
6. Push: git push origin feature/my-feature
7. Open a Pull Request with a clear description of what changed and why

Please open an issue first for large changes so the approach can be discussed before significant work is invested.

---

License

This project is licensed under the MIT License.