EarthCam has built connectors that let project teams ask ChatGPT, Claude and other AI assistants about their jobsites and get answers drawn from EarthCam’s camera feeds. The Upper Saddle River, N.J., company announced the construction AI connectors Oct. 20, along with seven new analytics models.
The connectors run on the Model Context Protocol, the open standard AI platforms use to plug in outside data. EarthCam’s version is listed as an app in OpenAI’s ChatGPT and as a connector in Anthropic’s Claude, and the company says it can be added to other MCP-compatible tools, including Google Gemini, Microsoft Copilot and Perplexity.
How EarthCam’s construction AI connectors work
The pitch is that a general-purpose model shouldn’t be the thing reading raw jobsite photos. EarthCam’s own models analyze the imagery first, sometimes on the camera itself, and hand structured results to the assistant. The company says that’s about ten times faster than having a customer’s AI agent interpret raw images, and it burns fewer usage credits. Answers link back to high-resolution images, timestamps and metadata so a superintendent can check them.
The data goes past video. The connectors tie into EarthCam’s Control Center 9 platform, including gigapixel panoramas from its Pano VI system, weather and IoT sensor feeds, and schedules, budgets and specs from connected systems. EarthCam also expanded schedule automation that compares planned milestones with what the cameras see.
Seven new jobsite AI models
The new analytics cover trench safety, housekeeping hazards, thermal imaging, delivery verification, dump truck analysis, visual occupancy and retail signage detection. Scott Travis, director of northern operations at GBR Roofing, said his team used EarthCam AI on a roof panel replacement to watch for fall-through hazards. Exclusion zones trigger alerts when someone walks in, so the crew can confirm fall protection is in place.
EarthCam says its models were trained on billions of construction images its cameras captured over three decades, on projects that include One World Trade Center, Hudson Yards and SoFi Stadium.
Checking the jobsite camera AI accuracy claims
The headline performance claim needs a label. EarthCam says that in its own benchmark testing, its AI identified construction objects with better context and accuracy, and no hallucinations, compared with leading general-purpose models looking at the same images. The release doesn’t publish the test set, the error rates or which models it compared. That’s the same pattern as Togal’s takeoff benchmark, which Togal wrote and graded itself. Pricing isn’t disclosed either. EarthCam points buyers to demos.
The design choice is the interesting part. Instead of selling one more dashboard, EarthCam is betting project teams will ask questions in whatever AI tool their company already pays for. If that holds, camera vendors end up competing on the quality of the structured data they hand over, and a claim like “no hallucinations” becomes something a GC can test against its own footage before it signs.