Proyecto TFG
Createdjaimealruiz
Interconnects Large Language Models (LLMs) with data spaces using the Model Context Protocol (MCP).
About
This project implements a functional and scalable architecture enabling an LLM to interact with a data space via the Model Context Protocol (MCP). It features a secure, modular MCP server mediating all LLM data access, preventing direct database interaction. The system includes an LLM client, a FastAPI-based MCP server connected to a DuckDB data space, and adheres to MCP principles for secure and extensible data interaction. The design emphasizes scalability for future integration with Apache Iceberg or Trino/Presto, advanced LLMs, and a RAG architecture.
Key Features
- Uses TinyLlama-1.1B-Chat-v1.0 for natural language queries
- Enriched prompts with contextual information from MCP
- Adherence to MCP design: LLMs access data only through tools
- Modular, extensible, and traceable architecture via logs
- Strict separation of semantic processing (LLM) and data access (MCP)
- 0 GitHub stars
Use Cases
- Building a foundation for Retrieval-Augmented Generation (RAG) systems
- Implementing a modular and extensible data access layer for LLMs
- Enabling LLMs to query and analyze data in a secure and controlled manner