MCP Model Context Protocol

🧩 MCP Agent with Web Search Integration

This project demonstrates how to build an intelligent AI assistant using LlamaIndex, FastMCP, and OpenAI's GPT-4o, enhanced with real-time web search capabilities via Serper API.

With this setup:

You can interact with an agent that answers your queries.

If needed, the agent fetches fresh data from the internet to provide accurate and up-to-date information.

Note:

This project uses uv (a faster Python package manager) instead of pip for dependency management.

Features

✅ AI Assistant powered by OpenAI gpt-4o

✅ Integrated with llama-index Function Agent

✅ Real-time Web Search using Serper.dev

✅ Async-first implementation for speed 🚀

✅ Custom MCP server and client communication

✅ Clean streaming of tool calls and responses

🧩 Tech Stack

• Python 3.11+

• uv (Python dependency management)

• FastMCP

• LlamaIndex

• OpenAI GPT-4o

• HTTPx (for async HTTP requests)

• Serper.dev API (Google Search API)

• BeautifulSoup (Web scraping)

Related MCP server: Perplexity MCP Server

📖 References

• Server-side code reference:
  MCP Quickstart Server

• Client-side code reference:
  LlamaIndex MCP Integration Example

🚀 Running the Project

1. Start MCP Server

uv run mcp_server.py --server_type=sse

2. Start MCP Client

python mcp_client.py

3. Usage Example

You: Who is the CEO of OpenAI?
Agent: Sam Altman is the CEO of OpenAI. (retrieved from latest web search)