Product · Assistants

Your register, on speaking terms with your assistant.

First-party MCP on every plan. Claude or ChatGPT can read your inventory - and, carefully, act on it. AI suggests; a human confirms; the ledger is never auto-written.

What's running low in Van 1?

Cable ties 300 mm - 4 on hand, minimum 5. Everything else in Van 1 is above its minimum.

Top it up by 20.

Here's what I'd do: adjust Cable ties 300 mm by +20 in Van 1 (4 → 24). Nothing is written yet - confirm and I'll post it to the ledger with your key on the record.

A real exchange shape: reads answer instantly; writes always show a preview and wait for a yes.

How it works

Connected in three minutes.

1
Create an API key
Settings → API keys, one click. A read-scoped key is free on every plan - including the free one - and it's all an assistant needs to answer questions.
2
Connect your assistant
Point Claude, ChatGPT, or anything MCP-speaking at api.itemtally.com/mcp with the key as the bearer token. No middleware, no Zapier, no CSV shuffling.
3
Ask in plain words
“What's low in Van 1?” “What does Sam have out?” “How long will the zip ties last?” Every answer comes from the movement ledger - the same numbers the app shows.
The rule that makes this safe

AI suggests. A human confirms.

Every number in ItemTally comes from an append-only movement ledger - that's what makes the register trustworthy. So the assistant rule is structural, not a policy: reads are free, writes are gated. A write needs a write-scoped key and an explicit confirmation after a preview. There is no configuration in which an AI quietly edits your stock.

What it can answer

The register questions, spoken.

find_items “Do we have…”, “how many…”, “where is…”
who_has_what “What does Sam have checked out?”
stock_levels “What's running low?”
days_of_stock_left “How long will the zip ties last?”
stale_stock “What hasn't moved in 90 days?”
inventory_summary Totals and value, by folder and tag
adjust_stock · check_out · check_in Writes - preview first, confirm second

Also inside the app: ⌘K

No connector needed - press ⌘K in ItemTally and choose “Ask the register”. Same tools, same ledger, same answers. Useful when you're already in the app and just want the sentence, not the spreadsheet.

Building something custom instead? The same capabilities are plain REST on the public API - every endpoint documented and live.

Honest answers

The questions people actually ask.

Can the assistant change my inventory?
Only if you let it, and never silently. Write actions need a write-scoped key (which counts as a crew seat), and every single action returns a preview first - it only executes when it's called again with an explicit confirm after you've agreed. With a read key, writes are refused outright.
Does this cost extra?
No. API keys - and the MCP server, which is the same API - are included on every plan, including the free one. Read keys are free and unlimited; a write key occupies a crew seat, same as a person who can write.
Is my data used to train AI models?
No. The MCP server answers your assistant's questions from your register at request time. Nothing is shared with anyone else's model, and nothing leaves the EU-hosted database except the answer to the question you asked.
What's the ⌘K thing in the app?
Ask ItemTally - the same plain-language questions, without leaving the app. Press ⌘K, choose “Ask the register”, and type. It uses the same read-only tools the MCP server exposes, so the answers always agree.
Which assistants work?
Anything that speaks MCP (Model Context Protocol): Claude, ChatGPT, and a growing list of clients and IDEs. If your tool can add a remote MCP server with a bearer token, it can read your register.

Ask your inventory a question today.

API keys are on every plan - including free. No card, no call.

Start free Next: Reports →

No card. No trial clock. Free forever for one person.