Web_Search_MCP
An MCP(Model Context Protocol) Server with a web search tool
This project demonstrates how to create a web search tool using the Tavily API and integrate it with MCP (Model Context Protocol) for seamless interaction with AI Systems. This allows you to provide real-time web search capabilities to your language models or applications.
Overview
The Web_Search_MCP project leverages the following key components:
- Tavily API: A powerful search API that provides real-time, comprehensive web search results, including answers, raw content, and relevant metadata.
- FastMCP: The FastMCP class uses Python type hints and docstrings to automatically generate tool definitions, making it easy to create and maintain MCP tools.
- Langchain: Specifically, the
TavilySearchResults tool from Langchain is used to interact with the Tavily API efficiently.
- Dotenv: A library for loading environment variables from a
.env file, securely managing sensitive information like API keys.
- uv: A very fast Python package installer and resolver, used to manage and run this project.
Functionality
The core of this project lies within the search_web tool, which provides the following features:
- Web Search: Accepts a search query as input and retrieves relevant search results from the web using the Tavily API.
- Detailed Results: Provides detailed information from the search results, including the website's content, URL, a relevancy score, the content type, and a direct answer (if available).
- Formatted Output: Returns the search results in a well-structured JSON format. The output includes a status indicator (
success or error), an array of results (if successful), and a timestamp.
- Error Handling: Gracefully handles errors during the search process and returns an informative error message in JSON format.
- Asynchronous processing : The search tool is based on asynchronous, which can handle many requests at the same time.
Prerequisites
Before running the project, ensure that you have:
Python 3.8+: Python 3.8 or a later version installed on your system.
Tavily API Key: A valid Tavily API key, obtainable by signing up on the Tavily website.
uv: The uv package manager for Python. You can install it using:
pip install uv
Installation
Create Project Directory: Create a directory for the project and navigate into it:
uv init Web_Search_MCP
*** By Running the above code uv creates pyproject.toml and .venv in the directory ***
2. Create Project Files: Create the files main.py and .env in the Web_Search_MCP directory.
3. Activate Venv: Navigate into .venv/Scripts/activate and activate the Vritual Environment
4. Copy code: copy the code in the main.py and .env into the files you just create.
5. Install Dependencies: Use uv to install the required Python packages:
uv add "mcp[cli]" python-dotenv langchain-community tavily-python
Configuration
.env File:
* Create a file named `.env` in the `Web_Search_MCP` directory.
* Add your Tavily API key to the `.env` file in the following format:
TAVILY_API_KEY='your_tavily_api_key'
Replace your_tavily_api_key with your actual Tavily API key.
2. claude_desktop_config.json:
* This file is used by the Claude desktop application (if you are using it) to discover and run the FastMCP server.
* It should reside in `c:\Users\<Your User Name>\AppData\Roaming\Claude\claude_desktop_config.json`
* Ensure the path to your `Web_Search_MCP` directory in `claude_desktop_config.json` is accurate. If your project is not in path/Web_Search_MCP`, please modify the`args` field in the config file.
{
"mcpServers": {
"Mcp_Demo": {
"command": "uv",
"args": [
"--directory",
"path/Web_Search_MCP",
"run",
"main.py"
]
}
}
}
Running the MCP Server
After Congfiguring the claude_desktop_config.json file, you can start the server from within the Claude desktop application.
Usage
Once the server is running, it exposes a single tool called search_web according to the Model Context Protocol.
- Tool Name:
search_web
- Input:
query (str): The search query you want to submit to the web.
- Output: A JSON formatted string containing the search results.
Example JSON Response (Success):
{
"status": "success",
"results": [
{
"title": "...",
"url": "...",
"content": "text",
"score": 0.9,
"raw_content": "..."
},
{
"title": "...",
"url": "...",
"content": "text",
"score": 0.9,
"raw_content": "..."
}
],
"timestamp": 1708849844.064655
}
