PMC, automated: replacing manual Excel scheduling
An end-to-end upgrade of the MRP process on top of Kingdee Cloud ERP. The system reads MRP work orders straight from Kingdee, schedules them against capacity and material readiness, and outputs shortage analysis and purchase suggestions — closing the loop from MRP work orders through scheduling and material analysis to the planning board.
Four pain points of manual production scheduling
The client already ran Kingdee Cloud ERP across sales, finance, inventory and material management, but the production planning and material planning modules were missing, so the PMC team still scheduled everything in Excel. This project plugs in as an ERP extension, replacing manual work with an automated loop from data capture to plan output.
- 01Manual process — schedules circulate as Excel files, planning data is consolidated by hand, and versions are nearly impossible to control
- 02Slow re-planning — every change in demand means recalculating from scratch, with repeated cross-checking that costs hours and invites errors
- 03Tedious kitting checks — subcontractor capacity has to be matched by hand, and every scheduling run needs round after round of confirmation
- 04Stale information — PMC, purchasing and management share no live board, so the plan travels as attachments
An automated loop from ERP data to the planning board
The system reads Kingdee MRP work orders, schedules them against capacity constraints and material readiness, and produces shortage analysis and purchase suggestions — one closed planning loop.
Where ERP ends and PMC begins
Kingdee keeps handling BOM explosion and supply-demand calculation. This system plugs in as an ERP extension that adds scheduling, shortage analysis and the planning board — without changing the ERP, it fills the gap the ERP left.
Scheduling · Shortage analysis · Planning board
Five stages implemented one by one in the system, each with a defined input and output.
- 1
DATA SYNC
Pulls sales orders, live inventory, BOM, in-transit purchases and work-in-progress orders from Kingdee via WebAPI, with retry, circuit breaking, rate limiting and a logging interceptor to keep the integration stable.
- 2
MRP WORK ORDERS
Kingdee handles BOM explosion and supply-demand calculation; this system reads the resulting PRD_Mo work orders as scheduling input instead of recalculating them.
- 3
PRODUCTION SCHEDULING
Runs in-house and subcontracted lines in parallel, weighing capacity, material readiness and due-date priority to generate a plan by line and by date.
- 4
SHORTAGE ANALYSIS
Aggregates material demand, nets it against stock, in-transit and safety stock, and generates purchase suggestions by supplier. Supports Excel export, offline manual adjustment, and bulk update on re-import.
- 5
PLAN DASHBOARD
PMC confirmation, purchasing review and management reporting all happen in one system, with every role sharing the same live view.
Where AI comes in
No new database was introduced: the AI layer reads entirely from the existing pp_ / mp_ / erp_ tables through a query-and-compute read layer, which keeps it safe and controllable. Scheduling is heuristic rather than mathematically optimal — a deliberate choice, because it matches how planners actually work.
MCP Server + natural language query (NLQ)
Embedded in Spring Boot and served over SSE, the MCP Server wraps Kingdee data queries, plan queries and scheduling operations as 24 tools that any MCP-compatible client — Claude Code, Cursor and others — can call. PMC staff and management can also ask questions in plain language and get answers from sales orders, live inventory, BOM, material plans and shortage analysis, without reports or query writing.
AI-assisted scheduling skills
Runs the workflow end to end: fetch unscheduled orders → apply constraints intelligently → write back the schedule → export a standard Excel file, respecting capacity, kitting and due-date rules.
MCP and NLQ: how the AI reaches real data
The AI is not a chatbot sitting outside the business systems. It reaches the real ERP data and business tools through an MCP Server, and hands the result back in plain language. That layer is controlled tool calling, not a direct database connection.
System Architecture
The stack combines a familiar enterprise application layer with a dedicated AI and MCP layer.
Delivery roadmap
The MVP already combines an automated loop with room for manual adjustment. V2 and V3 extend it on demand, which keeps delivery risk low.
MVP · 1-2 months
- Kingdee API integration
- MRP work order intake
- Production plan + shortage analysis + purchase suggestions
- Manual adjustment & report import/export
V2 · iteration
- Scheduled data sync
- Subcontractor capacity management
- AI capabilities (MCP / NLQ / scheduling)
- Notification system
V3 · optional
- Write back to Kingdee to create official documents
- Visual Gantt scheduling
- Supplier collaboration portal
Technology
Have a similar digital transformation need?
Let AI run in your real business scenario first. Book a free discussion about putting AI to work.