Supply Chain Control Tower: A Multi-Agent AI-Based Intelligent Decision-Making System for Supply Chains

Supply Chain Control Tower is an open-source multi-agent AI intelligent decision-making system for supply chains, built using the MCP protocol and Claude Desktop. It includes 12 professional agents and 57 tools to automate the entire process from order tracking to root cause analysis. The project aims to address pain points faced by modern supply chain teams such as fragmented systems and data silos, upgrading AI from a simple chat assistant to an operational decision support tool.

Modern supply chain teams often face issues like fragmented systems, manual spreadsheets, delayed reports, and data silos. This project demonstrates the practical application of multi-agent AI to address core needs such as operational visibility, root cause analysis, anomaly management, recommendation support, performance monitoring, and continuous improvement. The goal is to integrate AI into operational decision support workflows.

Technical Architecture

The system uses a layered design: Claude Desktop → Coordination Agent → Professional Supply Chain Agents → Shared Business Logic + SQLite Database → Structured Output/Dashboard/Logs. The coordination agent is responsible for routing user queries to professional agents.

Technology Stack

Component	Technology
Programming Language	Python 3.10
AI Protocol	MCP via FastMCP
AI Client	Claude Desktop
Fallback Model	OpenRouter
Database	SQLite
Dashboard	Streamlit + Plotly

Agent Ecosystem

It includes 12 professional agents such as freight delay analysis, inventory monitoring, purchase order tracking, root cause investigation, etc., providing a total of 57 tools covering multi-domain collaboration in supply chains.

1. Multi-Agent Reasoning: Claude collaborates with multiple agents to answer complex operational questions;
2. Supply Chain Intelligent Rules: Configurable rules for delay classification, inventory risk, carrier performance, etc.;
3. Continuous Improvement Layer: Detects repeated patterns and optimizes recommendations based on feedback;
4. Security-Aware Execution: Multi-layer security including read-only database access, input validation, prompt injection protection, etc.;
5. Performance Optimization: Query caching, token tracking, slow query detection, etc.;
6. Fallback LLM Support: Switches to fallback models when the primary model is unavailable to ensure continuity.

Typical Scenarios

Users can ask questions in natural language, such as:

• Which orders need to be processed today?
• Investigate the reason for the delay in SO10003?
• Which carriers are underperforming? The system automatically coordinates agents to generate structured answers.

Technical Contributions

Provides a reference architecture for MCP multi-agent applications, demonstrating 10 values including multi-agent orchestration, local AI workflows, domain-specific agents, lightweight data storage, centralized configuration, visual dashboards, secure execution, performance monitoring, automated testing, and continuous improvement.

Applicable Scenarios

• Supply chain operation teams seeking AI-assisted decision-making;
• Developers wanting to understand the practical application of the MCP protocol;
• Small and medium-sized enterprises needing a lightweight monitoring system;
• Researchers/students learning multi-agent architectures.

Limitations

• The demo version has limited data scale and functionality;
• Primarily aimed at education and proof of concept;
• Additional security and scalability considerations are needed for production environments.

This project demonstrates the application potential of multi-agent AI in the supply chain field. By connecting Claude with professional agents via the MCP protocol, it upgrades AI into a decision support tool. Core values include practicality, scalability, interpretability, and continuous learning. For teams exploring AI-driven operational decision-making, it is a valuable open-source implementation to reference.