Skip to content
Rush Commerce
Tools & Teardowns3 min read

Katana MCP: your inventory system now speaks agent

Katana shipped an MCP server and one-click AI replenishment for cloud inventory. Why a vendor-supported MCP endpoint matters more than the forecasting feature.

Most inventory software gets AI the same way: a chat box bolted to the corner of the dashboard that summarizes what you were already looking at. Katana MCP is the other approach — an endpoint that lets Claude or ChatGPT read and write your actual inventory records. Katana shipped it alongside a one-click replenishment forecaster, and the boring one is the important one.

What actually happened

Katana Cloud Inventory rolled out four things this month: Katana MCP, AI Replenishment, assisted spreadsheet import, and custom fields across modules.

Per Katana's own writeup, AI Replenishment reads your existing sales orders, manufacturing orders, POs, and stock — no CSV export step — and returns a 12-month demand forecast with three order scenarios (minimum, recommended, maximum), days-of-cover projections, safety stock, and a draft PO or MO at the end of it. You select SKUs individually or in batches of up to 50 and hit Replenish. It accepts plain-language context too: a competitor going out of stock, a marketing push landing next month.

The MCP server is the piece that changes the shape of the software. Per Sourcing Journal's coverage, it connects inventory operations to assistants including Claude and ChatGPT, and it's not read-only — it can create and update sales, manufacturing, and purchase orders. Katana describes it as one of the first vendor-supported MCP integrations in inventory management, which as far as we can tell is fair.

Why an MCP server matters for your business

A forecasting feature is a feature. Katana's is a good one — the three-scenario output with days-of-cover is more honest than a single number, and generating the draft PO at the end removes the step where a human retypes it wrong.

But a vendor-supported MCP endpoint is a different category of thing. It means your inventory system stops being a destination and becomes a tool that other software can call. Your reorder agent can check stock across Katana, pull open orders from Shopify, and cross-reference a supplier email — in one loop, without anyone writing a connector. That's the composition we normally build by hand against a REST API, and it took us weeks. Now it's a config block.

Two cautions, because the write path is real. First: an MCP server that can create purchase orders can create wrong purchase orders. Scope the token to what the agent actually needs and keep a human on the approve step for anything that spends money — we've made the approval gate argument before and it applies squarely here. Second: an agent is only as good as the data underneath. If your Katana counts are three days stale because someone stopped scanning, a 12-month forecast built on them is confident nonsense.

Fix the inputs. Then let the agents at it.

Key takeaways

  • Katana shipped an MCP server that lets Claude and ChatGPT read and write inventory, sales, manufacturing, and purchase orders
  • AI Replenishment returns a 12-month forecast, three order scenarios, days-of-cover, and a draft PO — up to 50 SKUs at a time
  • The MCP endpoint matters more: it turns the inventory system into something other software can compose with, no custom connector
  • Scope the token, keep a human approval gate on anything that spends money, and fix stale counts before trusting any forecast

Vendor shipped an MCP server and you're not sure what to do with it? We connect agents to the systems you already run — with scoped credentials and approval gates where the money moves. See how we build it or tell us what's in your stack.

Sources: Katana, Sourcing Journal.

  • #mcp
  • #inventory
  • #katana
  • #ai-agents
  • #erp
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.