ai_mcp_py — AI-Powered Web Search Assistant

A locally run AI assistant that uses Ollama (WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B or LLM of your choice) and scrape the web using Bright Data's Unlocker MCP tools — all orchestrated in Python and Streamlit with some Node.js under the hood. No cloud LLMs. No data sharing. Just local compute magic.

🔒 Local, Private, and Powerful — because your prompts are nobody else's business.

Built With

• Python + Asyncio
• LangChain
• Streamlit
• Bright Data MCP
• Ollama
• Terminal spinner magic courtesy of itertools

Setup

1. Get Bright Data API Key

• Go to brightdata.com
• Under Proxies and Scraping, create an Unlocker_MCP zone
• Make sure to:
  • Allow Admin Access
  • Set token to never expire
• Paste the token into your terminal session:

export BRD_API_KEY=your_token_here

💸 They offer free credits on signup (no credit card needed). Ignore the payment screens, click through.

2. Clone Repo

git clone https://github.com/drewesk/ai-mcp-py.git
cd ai-mcp-py

3. Create Conda Environment (Mac/Linux)

conda create -n mcp_env python=3.12
conda activate mcp_env
conda install nodejs

🧪 Make sure node, npx, and conda work from your terminal. You may need to update your .zshrc or .bashrc.

4. Install Python Dependencies

pip install -r requirements.txt

5. Start MCP Server

npx @brightdata/mcp API_TOKEN=$BRD_API_KEY

6. Start Ollama + Download Model

In a separate terminal:

ollama serve
ollama run WhiteRabbitNeo/WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B

Make sure the model runs smoothly on your machine (GPU preferred).

7. Run

streamlit run mcp_app.py

Example Usage

1. Paste URL → https://medium.com/ayuth/install-anaconda-on-macos-with-homebrew-c94437d63a37
2. Prompt → "What does this article tell me to do?"
3. Hit Submit, wait for the 🌀 spinner to complete
4. Voilà! — you'll see a local LLM-generated summary right on screen

Follow & Contribute

Open PRs, make suggestions, or fork at your liesure.