LuxePlatforms
Rather just talk?(816) 477-6802hello@luxeplatforms.comKansas City · Building nationwide
AI Tooling · ERP07 / 16

A production MCP server that runs an entire custom-millwork shop from a single chat window

Run a commercial millwork shop in plain English.

A Model Context Protocol server that gives plain-English control over a commercial millwork operation: quoting, submittals, change orders, deliveries, AR, retainage, and business intelligence. A general manager runs the business from one chat window with no command syntax.

In production · CI-tested

A look at the interface.

Representative UI — abstracted, never a client's real data.
🔒chat.workspace.dev
New chat
Recent
Quarterly report draft
Refactor billing logic
Summarize call notes
Schema migration plan
Workspace · Pro
AssistantonlineOpus 4.8 ⌄
Pull last quarter's numbers from the PDF and chart revenue vs target.
analyze_document(Q4-report.pdf)done
Revenue
$128k
Target
$110k
Δ
+16%
Revenue came in 16% above target. The lift is driven by Q4 enterprise renewals — here's the breakdown:
  • Enterprise +$22k (renewals)
  • Self-serve flat MoM
[1] Q4-report.pdf[2] forecast.xlsx
Ask anything…Opus 4.8
AI / conversational · representative UI
Talk
Run the whole shop by asking
Live
In production today
Forecasts
12-week cash flow, built in
Role
Solo, backend + AI integration
Year
2026
Status
In production · CI-tested
Category
AI Tooling · ERP
Built with
  • Python
  • MCP
  • SQLite
  • Pydantic
  • PyMuPDF
  • Bluebeam API
(01) The Challenge

Running a full shop conversationally and reliably meant 142 tools that tolerate missing data, stay type-checked, and pass CI on both Windows and Linux, with install profiles that degrade gracefully on hardware without build tools.

(02) The Approach

How it
works

142 tools cover quoting with win-rate prediction, materials with staleness alerts, shop production, and finance with a 12-week cash-flow forecast. On top of that sit 18 business-intelligence analyses. Read-only Resources expose every entity so the model reads state without spending a tool call, and per-tool safety hints plus a daily-rolled audit log keep destructive actions honest. SQLite with write-ahead logging is primary, with JSON mirrors as a read fallback.

What I built

(03) Inside the build
01

142 tools across 22 modules

Quoting with win-rate prediction, materials with staleness alerts, shop production and CNC queues, finance with AR aging and a 12-week cash-flow forecast, plus 18 business-intelligence analyses.

02

Resources, templates & prompts

19 read-only Resources expose every entity so the model reads state without spending a tool call, with ready-to-run workflow prompts like weekly review and architect chase.

03

Resilient, safe data layer

SQLite with write-ahead logging and JSON mirrors as a read fallback, plus per-tool safety hints and a daily-rolled audit log of every call.

(04) Highlights

The
receipts

  • 142 tools, 19 Resources, 9 templates, 7 prompts in one server
  • Deep Bluebeam integration: approval stamping, transmittals, CO drawing packages
  • Three install profiles for cross-platform deployment
  • Type-checked and CI-tested across Windows and Linux
← Previous
Physics-Simulated Cooking Game
Next →
AI Vision for Construction Drawings