Inspiration- HVAC intake is repetitive but mistakes can affect urgency, routing, and service decisions. We wanted to test whether WebMCP could give an AI agent structured tools while keeping every decision visible to a human.
What it does- The workbench contains fictional HVAC intake cases and exposes 9 WebMCP tools. An agent can inspect cases, classify trade/intent/urgency, identify missing information, and evaluate results with deterministic scoring.
How we built it- We built a browser-based evaluation workbench with structured WebMCP tools, a deterministic scorer, fictional test cases, and a human-visible interface so agent and human can work from the same information.
Challenges we ran into- The biggest challenge was making the tool behavior deterministic and easy to verify while getting WebMCP browser support working reliably.
Accomplishments that we're proud of- A live WebMCP-enabled application, 9 working tools, public source code, and repeatable scoring that makes agent decisions inspectable.
What we learned- WebMCP can reduce brittle browser automation by exposing structured actions directly to agents while preserving human oversight.
What's next for HVAC Intake Eval Workbench- Expand the benchmark, add more service-industry workflows, and test the same approach with real-world intake systems.
Log in or sign up for Devpost to join the conversation.