Weather MCP Agent (FastAPI + MCP SSE)

This project demonstrates a complete setup of a FastAPI-based MCP (Model Context Protocol) server paired with a FastAgent client using Server-Sent Events (SSE). It provides weather alerts and forecasts via the National Weather Service API.

Project Structure

.
├── README.md
├── agent.py                    # FastAgent client entrypoint
├── fastagent.config.yaml       # FastAgent configuration (MCP servers, agent name)
├── fastagent.secrets.yaml      # Secret keys (OpenAI, Anthropic, etc.)
├── pyproject.toml              # Project dependencies (setuptools/poetry)
├── requirements.txt            # (Optional) pip install list
├── src/
│   ├── __init__.py
│   ├── main.py                 # FastAPI server setup & SSE mount
│   ├── routes.py               # HTTP endpoints: /, /about, /status
│   ├── weather.py              # FastMCP server config & @mcp.tool functions
│   └── ...                     # egg-info, pycache, etc.
└── uv.lock                     # Lockfile for `uv` tool

---

1. Clone the Repository

Begin by cloning the repository to your local machine:

git clone git@github.com:sergeychernyakov/weather_mcp_agent.git
cd weather_mcp_agent

2. Create a Virtual Environment

It's recommended to use a virtual environment to manage dependencies:

python3 -m venv .venv

3. Activate the Virtual Environment

Activate the virtual environment before installing dependencies.

• Linux/MacOS:

  source .venv/bin/activate
  

• Windows:

  .venv\Scripts\activate
  

Installing Dependencies

Requires Python 3.12+ and uv or uvicorn.

# Using uv:
uv pip install -r pyproject.toml

# Or directly with pip:
pip install -r requirements.txt

If you don’t have a requirements.txt, install manually:

pip install fastapi httpx mcp[cli] unicorn fast-agent-mcp

---

Running the FastAPI MCP Server

Launch the server before starting the agent:

uvicorn src.main:app --reload

The server exposes:

• GET / → HTML welcome page
• GET /about → Plain text info
• GET /status → JSON status
• POST /messages → Internal endpoint for MCP tool calls
• GET /sse → SSE endpoint for MCP clients

---

Running the FastAgent Client

The agent.py script launches a FastAgent that connects to the MCP server via SSE:

# Ensure the server is running, then in a new shell:
uv run agent.py        # uses fastagent.config.yaml by default

# Or explicitly:
uv run agent.py -- --debug

---

Interactive Tool Usage

Once the agent prompt appears:

default >

You can invoke your weather tools directly:

# Fetch active alerts for Texas
!get_weather_alerts state="TX"

# Fetch a point forecast by coordinates
!get_weather_forecast latitude=29.76 longitude=-95.36

• The leading ! tells FastAgent to treat it as a tool call (not a chat).
• state="TX" must be in quotes; numeric args can be unquoted.
• Responses from the MCP server will stream back via SSE and print in your console.

---

Debugging & Development

• Inspect tools locally with MCP Inspector:

mcp dev ./src/weather.py

Opens UI at http://127.0.0.1:6274 and proxies SSE.

• Agent debug mode:

uv run agent.py -- --debug

Shows connection logs, SSE events, and tool calls.

---

🧑‍💻 Using with ChatGPT (Function Calling via OpenAPI)

You can use this server with ChatGPT custom GPTs or assistants that support OpenAPI schema-based tool calling.

1. Expose your server publicly

Run ngrok to expose your local FastAPI server:

ngrok http 8000

This will provide you with a public HTTPS URL like:

https://127c-156-253-249-23.ngrok-free.app

2. Import OpenAPI in GPT Builder

1. Open ChatGPT → Explore GPTs.
2. Choose your custom GPT or create a new one.
3. Under "Actions" → click "Import from URL".
4. Paste:

https://127c-156-253-249-23.ngrok-free.app/openapi.json

5. ChatGPT will fetch the schema and automatically detect:

• get_weather_alerts(state: string)
• get_weather_forecast(latitude: number, longitude: number)

6. Save and try asking:

What's the weather forecast in New York?

ChatGPT will invoke your FastAPI MCP server and return the result.

---

2. Function-Calling with the OpenAI API

If you’re writing your own ChatGPT-style client, just pass the OpenAPI schema to the ChatCompletion.create() call:

import openai
import json

with open("openapi.json") as f:
    spec = json.load(f)

response = openai.ChatCompletion.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Show me weather alerts for CA"}],
    functions=spec["paths"],
    function_call="auto",
)

The model will then invoke your /messages → /sse pipeline under the hood and return the tool’s JSON response.

---

Additional Notes

• The configuration file (fastagent.config.yaml) must match agent name and server keys.
• Ensure the weather MCP server is defined under mcp: in the config.
• fast-agent-mcp version ≥ 0.2.19 is required for SSE transport.

---

Author

Sergey Chernyakov
📬 Telegram: @AIBotsTech