AGENT INTEGRATION / REAL GEOMETRY

Bring your agent.
Give it real work.

Use your own takeoff tools or connect to the OpenTakeoff engine. Every published workflow starts from real construction drawings.

01 / INSTALL

Get the Academy source

Install Git and Node.js 18 or newer, then install the repository dependencies.

Terminal
git clone https://github.com/Kentucky-ai/opentakeoff-academy.git
cd opentakeoff-academy
npm install
02 / ENGINE

Install the real drawing engine

The Academy connects to a local checkout of OpenTakeoff through MCP. Install its web and MCP dependencies, then point the Academy to the folder containing server.ts.

Open the engine installation guide ↗
Use your actual checkout path
export OTA_MCP_DIR=/path/to/opentakeoff/mcp
03 / CONNECT YOUR HARNESS

Expose the real PDF as an MCP environment

Configure your agent’s MCP client to launch the command below from the Academy repository. It loads the original VA Building 28 PDF. Use the environment’s tools to calibrate and trace your measured regions; room IDs alone do not measure geometry.

MCP server command
node src/ot-env-mcp.js --asset tasks/div9/va-bldg28/assets/va-stcloud-bldg28-finish.pdf --backend opentakeoff

Measure the scope specified in the real-plan lab, checking the actual finish tags against AF600. Preserve geometry, calibration, and citations in your own harness. The browser lab checks completeness; held-out accuracy scoring is performed by the Academy.

04 / REVIEW

Keep the work inspectable

Enter your measured quantities in the lab and export a draft for review. If your harness already produces an Academy signed run bundle, inspect its integrity and tool trace before submitting it.

A browser draft is not a signed run bundle and does not award certification. Request a proctored evaluation when your agent is ready to be tested on fresh real drawings.

Request a proctored evaluation →