Nexus Chat: A Modular Intelligent Chat System and Practice of the MCP Protocol

Nexus Chat is a feature-rich modular intelligent chat system. Its core highlights include: support for multiple LLM providers (Anthropic Claude, OpenAI GPT, Ollama), built-in agent tool system and RAG (Retrieval-Augmented Generation) capabilities, full implementation of the Model Context Protocol (MCP) for standardized access to external services; the tech stack uses Python FastAPI for the backend and React for the frontend, with one-click deployment via Docker. It is one of the first open-source projects to fully implement the MCP protocol.

Nexus Chat integrates multiple LLM support, agent tool system, and RAG capabilities into one. It adopts a layered architecture design: the backend is built on Python FastAPI, the frontend is developed with React, and one-click deployment is achieved via Docker. Its core value lies in being one of the first open-source projects to fully implement the MCP protocol, demonstrating a standardized way for AI dialogue systems to connect with external services.

Multi-LLM Provider Support

Natively supports three major model families: Anthropic Claude, OpenAI GPT, and Ollama. You can flexibly choose based on scenarios (Claude for capability priority, GPT for cost priority, local Ollama for privacy priority). Configuration is managed via YAML, and API keys support environment variable injection.

Agent Tool System

Built-in common tools like calculator, code executor, and web search. Provides a low-threshold plugin extension mechanism (write a tool class → register configuration → import into the main module).

RAG Document Retrieval

Supports file upload and ChromaDB vector retrieval. Automatically performs document chunking, embedding, and indexing to enable context-aware Q&A.

Streaming Dialogue Experience

Real-time streaming responses are implemented via WebSocket. Users can view the model's thinking process word by word, enhancing the naturalness of interaction.

The Model Context Protocol (MCP) is an open protocol launched by Anthropic to standardize the interaction between AI models and external data sources/tools. Nexus Chat fully implements the MCP client, which can connect to any MCP server that complies with the protocol, and provides an SQLite MCP server example.

Connecting to a custom MCP service only requires three steps: 1. Implement the three standard methods: initialize, tools/list, tools/call; 2. Add a health check endpoint; 3. Register the service address in the configuration.

Layered Architecture

The code is layered by responsibility: providers encapsulate LLM calls, tools manage tools and plugins, mcp implements the protocol client, and rag handles document retrieval. Each module has clear responsibilities for easy maintenance.

Configuration-Driven Design

All variable behaviors (model parameters, tool switches, MCP connection information, etc.) are controlled via YAML configuration, allowing adaptation to different environments without modifying code.

Native Docker Support

Provides a complete Docker Compose configuration, supporting startup in standard mode and MCP mode. All services share the nexus-net bridge network to ensure smooth container communication.

Application Scenarios

• Enterprise internal unified multi-LLM management platform
• Vertical domain custom toolchain integration application
• Private document knowledge base Q&A system
• MCP protocol and tool integration experiment scenario

Development Experience

Provides cross-platform development scripts (start_dev.sh/bat) supporting Linux/macOS/Windows; the front-end and back-end separated architecture allows independent iteration, and WebSocket real-time feedback improves debugging efficiency.

Tech Stack

Backend: Python 3.11+, FastAPI Frontend: React 18+, Vite Vector Storage: ChromaDB Deployment: Docker, Docker Compose

Summary

Nexus Chat is a well-designed open-source chat system. It not only has practical functions such as multi-LLM support, tool system, and RAG, but also demonstrates a standardized integration method between AI systems and external services through the complete implementation of the MCP protocol, providing a highly referenceable sample for production-level AI application development.