import os
from openai import AzureOpenAI
from mcp.server.fastmcp import FastMCP
from internet_search import get_text_from_search
from dotenv import load_dotenv
load_dotenv(dotenv_path='./.env')
# IMPORTANT: If you plan on using an OpenAI client, or any other client the credentials have to be passed to instantiate the client.
# for demo purposes, we will use a persistent client tied to our environment variables.
openai_client = AzureOpenAI(
api_key=os.getenv('AZURE_API_KEY'),
api_version=os.getenv('VERSION'),
azure_endpoint=os.getenv('ENDPOINT'),
)
MODEL = os.getenv('MODEL')
# IMPORTANT: By default, the streamble-http transport layer is open w/ HTTP not HTTPS. Be careful when deploying
mcp = FastMCP("Demo", host='127.0.0.1', port=4000)
# IMPORTANT: The descriptions provided for the tools WILL be seen by the LLM when it begins to select a tool
@mcp.tool(description=
"""Adds two numbers"""
)
def add(a: int, b: int) -> int:
return a + b
@mcp.tool(description=
"""Tell's a family-friendly joke about the desired topic"""
)
def tell_joke(topic: str) -> str:
prompt = f"Tell me a short, family-friendly joke about {topic or 'anything'}."
response = openai_client.chat.completions.create(
model=MODEL,
messages=[{"role": "user", "content": prompt}],
temperature=0.8,
max_tokens=50
)
joke = response.choices[0].message.content.strip()
return joke
@mcp.tool(description=
"""
Searches the internet via Dux Distributed Global Search, and answers a user's question/query.
This tool will handle most of the user queries that require additional context.
"""
)
def search_internet_and_answer(query: str) -> str:
sys_prompt = f"""
INSTRUCTIONS:
You are a helpful AI assistant designed to answer users' questions using the context provided.
Your main audience is **Gen Z college students**, so keep your answers **full, clear, and casual**, feel free to use slang, memes, or relatable language.
Always answer **fully**, give examples if needed, and format your responses in **Markdown** for easy reading.
"""
# Step 1: Search the internet and gather the needed content
print("[1]: Fetching from Internet....")
internet_content = get_text_from_search(query)
# Step 2: Pass the context to the model
print(f"[2]: Passing Context...")
messages = [
{"role": "system", "content": sys_prompt},
{"role": "system", "content": f"CONTEXT: {internet_content}"},
{"role": "user", "content": query}
]
# Step 3: Invoke the model
print(f"[3]: Invoking LLM ...")
response = openai_client.chat.completions.create(
model=MODEL,
messages=messages
)
# Step 4: Send the final answer back to the client
print(f"[4]: Answer Generated, sending to client!")
answer = response.choices[0].message.content.strip()
return answer
if __name__ == "__main__":
rf"""
Run Command (From project-root directory): uv run .\mcp-server\main.py
"""
mcp.run(transport='streamable-http')
