How hsx-9506/mcp_test works
Project type and purpose
• This is an intelligent manufacturing data-preprocessing and multi-tool orchestration system built in Python.
• It uses a microservices (MCP) architecture to:
• Ingest raw Excel batch data at the edge (ETL)
• Cache structured JSON
• Expose multiple domain-specific APIs (KPI, SPC, anomaly, downtime, yield, etc.)
• Drive an LLM Agent that calls these services based on user intent and aggregates results.
File structure
• Top-level directories:
• agent_client/ ← Orchestrates calls to multiple MCP servers and feeds the LLM
• edge_etl/ ← ETL pipeline: Excel → JSON into json_cache/
• mcp_server/ ← One Python “server” per summary type reading from json_cache/
• intent_config/ ← Defines LLM intents and tool metadata (intents.json)
• config/ ← Global settings (setting.py, settings.json)
• mock_data/ ← Sample Excel data sets for testing
• Key patterns:
• Edge-processing → Cache → Microservices → Orchestration
• Each “*.py” under mcp_server is a standalone API endpoint
• Separation of concerns: ETL layer, storage layer, API layer, orchestration layer
Extract insights from README
• Workflow:
1. edge_etl/etl_to_json.py processes Excel → JSON cache
2. mcp_server/*_server.py each exposes a REST API for a different summary
3. agent_client/llm_agent.py uses intent_config/intents.json to break down questions, call servers, synthesize answers via LLM
• Dependencies: Python 3.9+, virtual environment, standard libs plus LLM SDK