GAD Editorial · Published · · English study · Chinese summary · Study dated 2026-10-02
A natural-language site-planning request contains more decisions than it first appears to. “Fit the housing and keep enough parking” leaves the unit target, parking basis, access requirements and acceptable compromises unresolved. TestFit's MCP connector makes that brief easier to send to software, but the handoff still deserves a reviewable specification.
The TestFit 5.25.0 release notes, dated 17 September 2026, add MCP operations including creating a deal, renaming sites and reading or setting floor-to-floor height. The vendor's FAQ describes a desktop-only connection between a chosen AI assistant and TestFit's deterministic engine. Availability in a particular office depends on its installed version and access. Release notes · MCP FAQ
Turn the request into explicit inputs
GAD recommends asking for a plain-language input summary before generation. List the project boundary, units, building type, quantity range and each fixed assumption. Identify preferences separately from hard constraints. Where a phrase admits several interpretations, resolve it before the assistant submits the study.
Consider an illustrative request for a residential option with a target unit count and a parking ratio. The team should specify what that ratio measures and which spaces count toward it. It should also identify which access and landscape requirements must remain. The assistant's summary can expose missing decisions while the project is still easy to correct.
Save a baseline copy and record the source of important inputs. A zoning assumption, a client preference and a survey dimension need different provenance. An internally consistent solve cannot establish that all three inputs are current or appropriate.
Inspect the actual handoff
After an operation, compare the software's displayed parameters with the agreed input summary. Pay particular attention to units, lower and upper bounds, and the difference between floor count and floor-to-floor height. Ask the responsible designer to inspect the resulting plan and its key quantities before another iteration begins.
Change one consequential condition at a time when the purpose is to understand an effect. If the team relaxes parking and changes height together, label that as a combined scenario. Preserve the earlier result so a reviewer can explain why the preferred option improved.
The deterministic engine described by TestFit offers a useful distinction: repeatable computation still depends on receiving the intended inputs. Treat errors in interpretation and errors in the resulting plan as separate review categories.
Keep manual work in the record
TestFit documents working between direct editing and MCP prompts. After a manual adjustment, have the next request refer to the current state and verify that the adjustment remains where required. Do not rely on the conversation alone to establish which version the team approved. Getting Started
The same guide says the desktop application and agent must be open, and that site-credit consumption requires manual confirmation. Make those operational boundaries visible before a demonstration. End with the saved study, its accepted constraints and a short list of unresolved checks. GAD has not operated the connector or measured its speed; this is a documentation-based workflow for making a conversational planning session easier to audit.