NeuroCube

Model Context Protocol (MCP) Integration

NeuroCube is evolving to integrate the Model Context Protocol (MCP) for better tool calling ( >ᴗ< ).

What is MCP?

MCP is a protocol designed to standardize how models interact with tools, allowing for more robust and flexible tool usage. It enables models to call tools in a structured way, improving the reliability and interpretability of interactions.

MCP Before and After MCP Architecture

MCP Integration

I’ve integrated MCP into NeuroCube with the help of langchain-mcp-adapters.

Retrieval Augmented Generation (RAG) tool

The RAG tool allows the model to retrieve and generate more relevant information by accessing external knowledge sources while also generating contextually relevant responses. I am using Milvus as the vector database to store and retrieve relevant information.

AI Search tool

The AI Search tool enables the model to retrieve information from the internet using AI Summaries, enhancing its ability to answer questions and provide up-to-date information. I’ve switched from using DuckDuckGo to using the Gemini API for more structured and accurate search results.

Next Steps

  1. Digital twin
    • The digital twin is still in the early stages of development and is largely simplified.
    • Next Steps: Improve digital twin to better reflect the real robot and eventually use it for simulation and testing.
  2. Revise locomotion and navigation system
    • The robot can only perform one action at a time for each human voice command. It also cannot perform actions on its own to solve a more complex task such as navigation.
    • Next Steps: Implement a more sophisticated navigation system that allows the robot to perform multiple actions in sequence and navigate to specific locations autonomously.