AI Engineering Drawing Estimator v0.1.2

AI Agent MCP for Symbol Counting and Quantity Takeoff

An assisted-review MCP server and Agent Skill for counting electrical and communication symbols in engineering PDF drawings with ChatGPT Codex, Hermes, and other MCP-compatible AI agents.

This is not a fully automatic quantity takeoff system. Every candidate must be reviewed before reporting a final count.

Supported workflow

• Inspect vector, hybrid, and raster PDF pages.

• Render pages for visual review.

• Detect vector candidates for:

  • Duplex Socket Outlet

  • Single Socket Outlet

  • Data Outlet

• Suppress candidates whose center overlaps PDF text by default.

• Create project-specific symbol templates from a clean legend ROI.

• Export candidate crops, markup, CSV, JSON, and an HTML review page.

• Confirm reviewed candidates and add manually verified points.

No project drawings, customer files, marked-up plans, test outputs, or machine-specific paths are included in this repository.

Related MCP server: PDF2MD MCP Server

Requirements

• Windows

• Python 3.12-3.14

• PowerShell

• An MCP-compatible AI agent such as Codex or Hermes

Install

Clone or download this repository, then run:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
.\install.ps1

The installer:

1. Creates .venv and installs the Python dependencies.

2. Runs the MCP doctor check.

3. Installs the Skill to ~/.codex/skills/count-engineering-drawing-symbols by default.

4. Generates codex-mcp-config.toml with the local MCP command.

Add the generated MCP block to ~/.codex/config.toml, then restart the agent.

When upgrading from v0.1.1, remove the old [mcp_servers.drawing-estimate-reader] block and use the generated [mcp_servers.engineering-drawing-estimator] block.

For another agent, register a stdio MCP server using:

command: .venv\Scripts\python.exe
args:    mcp\server.py

Agent prompt

Use $count-engineering-drawing-symbols to inspect this PDF.
Count Duplex Socket Outlet, Single Socket Outlet, and Data Outlet by floor.
Create markup and review every candidate before reporting the quantities.

MCP tools

Important review rules

• Treat starter templates as search aids, not universal standards.

• Prefer the legend from the current project.

• Check every wall face, wall corner, and both sides of doors.

• Support rotated symbols at 0, 90, 180, and 270 degrees.

• Use OCR only for text context, never for detecting symbol geometry.

• A nearby 4 can be a circuit number; C can identify CCTV.

• A zero result remains provisional until the candidate set and wall/door regions are reviewed.

• Use manual_points for visually confirmed misses.

Limitations

• The bundled matcher is intended mainly for vector/hybrid PDFs.

• Raster scans require high-resolution tiling and a vision/object-detection workflow.

• Different companies and projects may use different symbol geometry.

• Final results require human or agent-assisted visual confirmation.

License

MIT