hvac-diagnostic

HVAC symptom diagnosis

Guide a technician from a described symptom to a ranked, sourced diagnosis with a confidence score.

When to use

The user describes a behaviour or fault ("no hot water", "outdoor unit iced up", "défaut débit d'eau") rather than a specific error code, and wants to know the likely cause and next checks.

Prerequisites

The AskMarcel MCP server must be connected (https://mcp.askmarcel.app). If not, point the user to https://app.askmarcel.app/developers.

Steps

1. Gather context: symptom description, and if available brand and model. Ask once for brand/model if missing — it sharply improves accuracy.
2. Diagnose:
   • If you have a brand and an exact SKU (nameplate), call diagnose_v2. If status is not found, tell the technician there is no exact manual — do not invent a procedure from a family document.
   • Otherwise call diagnose with { query: "<symptom>", brand?, model? }.
   • diagnose returns ranked causes, steps, a confidence score, and source citations. diagnose_v2 uses the same shape with status and a nullable diagnostic.
3. Deepen where needed:
   • For any procedure referenced, call get_procedure with its chunk_id.
   • If the diagnosis points to an error code, call get_error_code to confirm.
   • For specs (refrigerant, pressures, tolerances), call get_product_sheet with { brand, model }.
4. Present:
   • Most-likely cause first, with the confidence score.
   • Ordered checks/actions.
   • Sources (manual + page). Offer get_pdf_page_snapshot for the exact page.
5. Safety & honesty: surface the confidence score. If low, say so and list what additional info (brand, model, measurements) would improve it. Never present an unsourced guess as fact.

Example

User: "Atlantic Alfea heat pump, water flow fault, won't run."

• diagnose { query: "water flow fault", brand: "Atlantic", model: "Alfea Excellia" }
• Return ranked causes (circulator, air in circuit, flow switch, filter…), the checks in order, confidence score, and cited pages.