Jeremy Neely Text Jeremy

AI & Agents · Beta

CAD Drawing Index

Lets staff ask their AI assistant what is inside AutoCAD drawings.

Built for: A manufacturing and wholesale company with its own engineering and drafting staff

Asking the AI assistant what is inside a set of drawings (sample data).
Asking the AI assistant what is inside a set of drawings (sample data).

The problem

The client's AI platform could find CAD files by name and date but could not see inside a drawing. Questions like "where is this tag shown?" or "count every fixture type in this building" meant opening drawings one at a time in AutoCAD.

What I built

I built an indexer that reads every .dwg and .dxf file on the file share, extracts blocks, quantities, attributes, layers and text into a database, and rescans changed files nightly. A read-only MCP server gives the client's AI assistant 12 query tools, so staff can ask plain-English questions and get counts and file paths in seconds.

How it works

  1. Crawl the drawings share with a read-only account.
  2. Convert and read each drawing's blocks, attributes, layers and text.
  3. Store it in MariaDB with each file's original path.
  4. Rescan only new or changed drawings every night.
  5. Answer the AI assistant's questions through 12 read-only MCP tools.

Built with

  • Python
  • ezdxf
  • MariaDB
  • MCP Python SDK
  • Docker Compose
  • Hatz AI

More of Jeremy's work