AI / Agent access (MCP)
AI / Agent access lets you connect your own AI assistant (Claude (desktop, web, or CLI), ChatGPT, or a custom agent) to your Logistified data over MCP (the Model Context Protocol). Once connected, you can ask your assistant questions about your purchase orders, suppliers, inventory levels, stock takes, inventory plans, and more, in plain language, and it reads the answer straight from Logistified.
In the Logistified admin this is the “LLM / Agent Access” tile inside Cloud PlatformCloud PlatformA separate, browser-based interface for warehouse devices: tablets, phones, handheld scanners. Logs in via a session link generated from inside Logistified. Enhanced-plan feature. Read more → (top nav → P&H Cloud Platform (Beta) → LLM / Agent Access).
On this page
Section titled “On this page”- What you can do with it
- How it works (code mode)
- Read-only by default
- Troubleshooting
- Where to next
What you can do with it
Section titled “What you can do with it”Connect an AI assistant and ask things like:
- “Which open purchase orders are arriving in the next two weeks?”
- “List my suppliers with the longest lead times.”
- “What’s the on-hand quantity for SKU ABC-123 across all locations?”
- “Summarise my most recent stock take.”
- “Which products in my main forecast view are about to run out, and how much should I reorder?”
- “How is my spring sales plan doing against actual sales at each store?”
The assistant reads live data from your store and answers from it: no copy-pasting, no exports.
How it works (code mode)
Section titled “How it works (code mode)”Logistified exposes a single, powerful tool to your assistant called run. Instead of the assistant chaining dozens of small requests, it writes a short program against a typed app API and Logistified runs it once, safely, in an isolated sandbox:
run("return await app.purchaseOrders.list({ status: 'open' });");This “code mode” approach lets the assistant filter, join, and summarise your data in one step (for example, listing open POs, joining them to supplier names, and sorting by arrival date) without you writing any code yourself. You just ask in plain language; the assistant writes the program.
Your assistant can also read this very documentation site through two helper tools (list_docs and get_doc), so it can ground its answers in how Logistified actually works. For common jobs such as a sales question, a weekly store report or reordering, it can load one of Logistified’s ready-made recipes (list_skills and get_skill) and follow its steps.
Read-only by default
Section titled “Read-only by default”By default, AI / Agent access is read-only: your assistant can look at your data but cannot change anything. If you explicitly grant write access (a write-scope API key, or a web chat connection granted write when it’s linked), the assistant can also operate your purchase orders, transfer orders, stock takes, supplier returns (RMAs), and reorder rules, your supplier directory and cost mappings, scheduled reports, notification rules, and email recipients, your saved forecast views, your Data Modeling tables and Product & variant attributes, your inventory plans (commitments, sales plans and launch plans), promotional events and planning options, and your purchase-order Constraint Studio rules and optimizer. Locations and sales orders remain read-only, and your Shopify product catalogue itself can’t be edited; the assistant can only write Logistified’s own per-variant planning fields (forecast settings, unit costs). See the operating pages below for what each write surface covers.
Troubleshooting
Section titled “Troubleshooting”The assistant says calls succeed but return no data
Section titled “The assistant says calls succeed but return no data”If your assistant reports that every request “succeeds” but comes back empty, even for things you know exist, the usual cause is how it read the result, not an outage. Each run answer is wrapped as { ok, data, requestId }, but inside the program the assistant writes, Logistified hands back the data directly. An assistant that looks for an extra data layer inside its program finds nothing and reports an empty list.
Ask your assistant to call describe_api for the method it is using (for example describe_api with that one method selected), check its return type, and run the request again without the extra .data. For example: “Re-run that, reading result.purchaseOrders directly instead of result.data.” If the answer is still empty, send the request IDs from the responses to support so we can trace them.
A long answer ends with ”--- TRUNCATED ---”
Section titled “A long answer ends with ”--- TRUNCATED ---””To keep your assistant’s working memory free for your question, Logistified trims very large answers (roughly 24,000 characters) before handing them over. It never cuts a record in half: whole records are kept or left out, and every cut is marked --- TRUNCATED --- with a count of what was left out. Page markers such as the next-page cursor and the total count always stay intact.
A capable assistant handles this on its own, by fetching the next page or asking for fewer fields. You can also ask directly: “That was cut off. Fetch the next page” or “Return only the PO number and status.” When it only read data, the assistant may repeat the request with trimming turned off. After a change, it should never repeat the change itself just to see the full answer; it reads the result back instead.
A request takes too long and times out
Section titled “A request takes too long and times out”Each program your assistant runs has a 30-second execution limit. A question that needs several reads can reach that limit before returning an answer. Ask for one small page or a smaller group of results at a time, with the assistant keeping each returned answer and its place before continuing. For example: “Return one page of purchase orders, keep that answer, then ask for the next page.” Smaller requests leave more time to finish; they cannot guarantee completion.
A timed-out request does not return the information it collected before failing. Answers already returned by earlier requests can still be kept. For read-only work, the assistant should follow the response’s wait-and-retry guidance and try a smaller request once, one request at a time.
If the request could make changes, a timeout does not prove those changes failed. Work already started may still finish. Ask the assistant to check what was saved before deciding how to continue.
A request failed partway through. Is it safe to try again?
Section titled “A request failed partway through. Is it safe to try again?”An assistant’s request can make several changes in a row. If it fails partway after starting, Logistified reports what it knows about the attempted work and whether repeating the whole request is safe. That report can be incomplete, and unfinished work may still complete. If a change was already made (a purchase order created, a line updated), repeating it could do the change twice, for example create a second purchase order. A good assistant checks what was already saved and continues from there. If yours offers to “just run it again”, ask it to check first: “Check whether the draft PO already exists before retrying.”
Where to next
Section titled “Where to next”