Inventory: Natural Language Access to Household Items via AI

"Where Did We Put the Extension Cords?"

That question cost me twenty minutes last week. I knew we had them. I knew they were somewhere in the front porch storage area. But the front porch has six bins, two shelves, and a cabinet, and I opened all of them before finding the cords in the last place I looked — because of course.

We have a household inventory system called Quartermaster, backed by Notion, that tracks items and their locations. I built it precisely to prevent this kind of scavenger hunt. The problem was that nobody actually opened Notion in the moment. When you're standing in front of six bins, you're not going to unlock your phone, navigate to a database, construct a filter, and wait for results. You're going to open bins.

What we needed was simpler. We needed to just ask.

What the Skill Does

The inventory skill wraps the Quartermaster MCP server with natural language understanding. It triggers on the kinds of questions you actually ask out loud:

• "Where is the tape measure?"
• "What's in the staff room?"
• "Do I have any extra USB cables?"
• "Find my passport"

Behind the scenes, it calls MCP tools to search items by name, filter by location or category, and return structured results including condition, photos, and the full location hierarchy (room → cabinet → shelf → bin). The answer comes back in seconds, in plain language, without touching Notion.

Beyond Search

The skill is not read-only. It handles the full lifecycle.

Adding items. "Add a new item: blue tarp, in the front porch storage bin" creates the item with proper location relations. I use this when unpacking deliveries — adding things as I put them away means the database stays current without a separate data-entry step.

Updating. "Move the drill to the workshop shelf" updates the location link. Things move around a house constantly. If updating the database isn't as easy as moving the object, the database rots.

Archiving. Soft-delete for items you've given away or tossed. The history stays without cluttering search results. In Indonesia, we give things away a lot — clothing, kitchen supplies, kids' toys the children have outgrown. Keeping a record of what left the house and when turns out to be useful.

Image identification. Take a photo of something and the AI vision tool identifies it against your inventory. This solves the problem of "what is this thing and where does it go?" — especially useful when Nurul finds something in the wrong place and wants to return it without asking me.

Label printing. Generate and print thermal labels for storage locations, so the physical bins match the database. This ties back to the thermal printer setup — the inventory system and the labeling system reinforce each other.

Location Hierarchy

Quartermaster models locations as a parent-child tree. A query like "what's in the living room?" returns items in the room itself plus everything in sub-locations — the TV cabinet, the bookshelf, each drawer. The skill understands this hierarchy and presents results grouped by sub-location rather than as a flat list. The difference matters: a flat list of 40 items is noise, but the same items organized by where they sit is immediately useful.

The Boundary

One important design choice: the skill explicitly does not trigger when you're asking about the Quartermaster codebase itself. "How does the inventory API work?" is a development question; "where is my inventory of screws?" is a usage question. The trigger phrases are tuned to catch the latter and ignore the former. Without this boundary, the skill would hijack unrelated coding conversations every time someone mentioned the word "inventory."

The real test of this skill is not whether it works — it does — but whether people use it instead of opening cabinets. So far, the answer is yes, mostly because asking is faster than standing up. Laziness, it turns out, is the best driver of adoption.