🧠 Re-Act Agent with MCP Brave Search Integration

This Jupyter Notebook demonstrates how to use the BeeAI Re-Act Agent framework integrated with the Model Context Protocol (MCP) Brave Server to perform intelligent internet searches and local lookups via Brave’s Search APIs.

---

🚀 What It Does

• Connects to the Brave MCP server using a local subprocess.
• Discovers tools (brave_web_search, brave_local_search) via MCP.
• Initializes a Re-Act Agent (based on OpenAI GPT-4o) that can:
  • Decide when to use each tool.
  • React to responses and iterate intelligently.
• Runs a conversation loop allowing user queries and AI responses.

---

🛠️ Setup

1. Install dependencies: Make sure you have:

   • beeai-framework
   • mcp
   • dotenv
   • Access to npx and the Brave MCP server package

2. Set up .env file:

3. Adjust paths:

• Update the .env path and any system paths according to your local setup.

---

🧪 Run the Notebook

1. Load the environment and dependencies.
2. Start the Brave MCP server and discover available tools.
3. Initialize the Re-Act Agent with LLM, memory, and tools.
4. Run the agent in a conversation loop to handle user queries.

---

🐞 Notes

• There's a known bug in the Brave MCP server requiring a count=5 parameter for certain queries. This is temporarily patched via the system prompt.
• Outputs are color-coded and printed via a helper ConsoleReader class for easier readability.

---

🌐 Tools Used

• brave_web_search: Web searches with filtering/pagination.
• brave_local_search: Local business lookups with detailed metadata.

---

✅ Example Use Case

"What are the opening hours for La Taqueria on Mission St in San Francisco?"
→ The agent will automatically select the appropriate tool and return real-time business details.