Claude-Powered Web Search Interface

!image

A minimal and scalable AI-powered search engine built using Claude's API and DuckDuckGo, integrated with both a web interface and a CLI tool.

---

📦 Installation

1. Clone the Repository:

   git clone https://github.com/your-username/claude-search.git
   cd claude-search
   

2. Create a Virtual Environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install Dependencies:

   pip install -r requirements.txt
   

4. Set Environment Variables:

   set CLAUDE_API_KEY=your_claude_api_key 
   # In powershell
   $env:CLAUDE_API_KEY=your_key
   

---

🛠 How It Works

This project includes two main interfaces:

1. 📟 CLI Interface (ask_claude.py)

• Sends queries to Claude.
• Claude detects tool-use (e.g., web search), and calls the MCP server to fetch DuckDuckGo results.
• Claude summarizes the results and replies.

Run:

# Start MCP Server
python mcp_server.py
# On another terminal run 
python ask_claude.py "What is the latest on AI governance?"

2. 🌐 Web Interface (app.py)

• Simple HTML UI for user-friendly searches.
• Sends a query to your MCP server (/tool_call) which fetches relevant results.
• Displays them in a clean, TailwindCSS-powered interface.

Run:

python app.py

Then visit: http://localhost:5001

---

⚙️ MCP Server (mcp_server.py)

The MCP (Modular Command Processing) server:

• Routes tool calls from Claude to actual data providers like DuckDuckGo.
• Responds with results which are then used by Claude or your app.

Endpoints:

• /health: Check if the MCP server is running.
• /tool_call: Accepts tool call requests with a query and returns relevant search results.

---

🚫 Why It Might Not Work Now

While the app and CLI are configured correctly and successfully connect to:

• ✅ Claude's API endpoint
• ✅ MCP server running locally

You may still receive error messages like:

"Your Claude API quota or credits may be exhausted."

!image

This is because Claude is rejecting requests due to exhausted API usage quota. The backend is still functional and integrated correctly.

---

📁 Project Structure

.
├── app.py              # Flask server for the website
├── ask_claude.py       # CLI interface
├── claude_mcp_client.py
├── mcp_server.py       # Handles Claude tool calls
├── mcp_integration.py  # Bridge to DuckDuckGo & Claude logic
├── static/
│   └── index.html      # Frontend UI
├── requirements.txt
└── README.md           # You're here!

---

📌 To Do

• Add Claude response summarization to UI.
• Enable query history per session.
• Add fallback to local GPT models when Claude is down.

---

🧠 Credits

Developed with love and frustration by \Rahul Agarwal. Uses Anthropic Claude API + DuckDuckGo search.