Product · AI agents · Model Context Protocol
InfiPlex ships a native MCP server on every instance. Connect Claude, ChatGPT (through Codex), Cursor, or any MCP compatible client to your own yourcompany.infiplex.com, and your AI assistant can search orders, explain why an order shipped from a specific warehouse, check inventory across warehouses, list what needs restocking, record tracking, set prices, and read or change settings. Every change to stock, prices, shipments, products, warehouses, or settings is previewed before it happens, and every tool is gated by the API key scopes you already control.
See it work
This is what an AI assistant connected to an InfiPlex instance actually does. You type a question. The assistant picks the tools, chains them, and shows you the preview before any write. Pick a scenario to replay it.
Nothing in the demo is invented. order_warehouse_routing_explanation_get returns the exact routing event log InfiPlex wrote when the order was allocated. inventory_restock_list uses the same weighted reorder point as the Restock Alerts email. price_set stops at the preview until the token comes back. marketplace_connect_link_get is how 3PLs already onboard client marketplace accounts, now callable by an agent. log_search reads the same shop log your support team reads, so "why did this not import" is answered from the record.
Headless order management
InfiPlex has always been the engine behind the screens: it pulls orders from 130+ sales channels and retailers, routes them across every warehouse, syncs inventory back out, and pushes to NetSuite, QuickBooks Online, 3PLs, and EDI retailers. The MCP server makes that engine callable by the AI agent your team already talks to, on a key whose permissions you set, from read-only to fully autonomous. That is headless order management: the engine keeps running, and the interface is whichever assistant you point at it. Connecting takes less time than reading this paragraph: five-minute set-up.
What you can ask
The assistant is not guessing from a knowledge base. Every answer is a live call into your InfiPlex instance through one or more of the tools. Here are the questions each group of tools answers, with the tool names so you can find them in the full tools reference.
Find, read, annotate, ship, and close orders from any channel.
order_searchorder_getorder_warehouse_routing_explanation_getorder_tracking_createorder_note_createorder_additional_info_createorder_processed · "Cancel it." order_cancel · "Move it to needs review." order_update_statusQuantities per warehouse, history, restock, bulk adjustments.
inventory_getinventory_searchinventory_change_history_getinventory_restock_listinventory_setinventory_set_bulkbin_location_setRead and change what customers see, with a preview first.
price_get · "Which prices changed this week?" price_searchprice_set (preview, then confirm)products_createpim_sku_getpim_nightly_fileThe structure your multi-warehouse routing runs on.
warehouse_search · "Show me Reno's routing settings." warehouse_getwarehouse_setwarehouse_createDiscover, read, and (if you allow it) change account behavior.
shop_setting_searchshop_setting_getshop_setting_setCredential settings are never returned and cannot be written. Settings that start live marketplace traffic need arming enabled on the key.
Built for 3PLs onboarding client accounts.
marketplace_connect_link_getReturns a one time link the marketplace account owner clicks to authorize, no InfiPlex login needed on their side. Expires the same day. Covers Amazon Selling Partner, Walmart, Etsy, TikTok Shop, and Shopify push and pull.
The shop log InfiPlex writes for every channel call and routing pass, readable by the assistant.
log_types_getlog_search on Amazon Get Orders: Failedlog_search with the order idEvery channel has failed, successful, and Any Log types. The assistant narrows by type and date, then pairs the entry with order_get or shop_setting_get to explain it.
Write safety
The reason most operations teams will not let an AI near their orders is that a wrong write propagates. An inventory set goes to every listing. A price change goes to every channel. A cancellation goes to the marketplace and cannot be undone. The InfiPlex MCP server is designed around that reality with four independent layers.
inventory_set, inventory_set_bulk, and bin_location_set are gone from the tool list.shop_setting_set is refused unless MCP Allow Setting Writes is also enabled for that key. Reads still work. Credential settings cannot be read or written through MCP at all.Five low-impact writes (internal notes, the processed flag, order status, additional info fields, and bin locations) execute on the first call because none of them changes a quantity, price, or shipment; they remain limited by the key's write scopes. Everything an agent does runs through the same audited API key path as a normal REST call, so your existing API logs show the agent's activity by key. The tool descriptions themselves instruct the assistant to read before writing (price_get before price_set, warehouse_get before warehouse_set) and to never re-enter an order that already exists on a channel. The write safety guide covers each layer with the exact refusal messages.
Who this is for
The person who owns Amazon, Walmart, Temu, Shopify, and TikTok Shop asks one assistant instead of opening five seller centrals. "Anything unshipped and past its Temu ship-by?" is one question. Inventory questions resolve across all warehouses in one call, and the assistant can explain a low quantity from the change history rather than guessing.
Onboarding a client's marketplace account becomes "generate a Walmart connect link for Acme" and forwarding the URL. Receiving becomes a spreadsheet applied through inventory_set_bulk with a preview of every row. Per-client API keys mean each client's agent, or your own, sees only that client's instance. See the 3PL Connector.
The additional_info fields that drive the NetSuite push (netsuite_class_id, netsuite_location_id, any netsuite_custom_field_*) are readable and settable on an order before it pushes. The routing explanation tool answers the "why did this hit the wrong location" question in seconds, with the rule that fired. The field mapping and order rules guide lists every field name the agent can set.
Setup
In your InfiPlex admin, open an API key (or create one), set MCP Enabled to Yes, choose the scopes you want the agent to have, and save. The key page shows the two values every client needs.
Settings > Connectors > Add custom connector. Paste the URL, leave OAuth blank, add the api-token header. On Team and Enterprise plans an Owner adds it under Organization settings, then members click Connect.
Any Streamable HTTP MCP client. Send the key as api-token or Authorization: Bearer. Both are accepted.
Step by step with screenshots for each client: MCP server set-up guide. Ready-to-use prompts for common operations: AI agent prompts for order management.
Why a native MCP server
There is a difference between a platform bolting a chat box onto its dashboard and a platform exposing its operations as tools an AI can call. A chat box answers what the vendor trained it to answer, inside the vendor's UI, with the vendor's model. An MCP server hands the tools to the assistant you already use, in the chat window you already have open, with your company's model and data policies. InfiPlex chose the second path.
Practically, that means three things. The endpoint is on your own subdomain, so requests never pass through a shared InfiPlex AI service. Authentication is the API key you already issue and revoke, with the same scopes. And because the server implements the Model Context Protocol rather than a proprietary plugin format, it works today with Claude, Codex, Cursor, and every client that adopts the standard next, with no InfiPlex release required. The REST API is still there for code; the API overview covers both.
Frequently asked
MCP (Model Context Protocol) is an open standard that lets an AI assistant call tools on an external system. An MCP server for order management exposes operations like search orders, get inventory, record tracking, and set price as tools with typed inputs, so an assistant such as Claude or ChatGPT can run them against your live data in response to a plain-English request. InfiPlex runs one on every instance at yourcompany.infiplex.com/mcp.
Claude on claude.ai and Claude Desktop (as a custom connector), Claude Code, ChatGPT through the Codex CLI and IDE extension, Cursor, Windsurf, and any other client that supports remote MCP servers over Streamable HTTP with a custom header or bearer token. Clients that only support OAuth login for remote servers are not supported at this time.
No. Every tool that changes stock, a price, a shipment, a product, a warehouse, or a setting requires two calls. The first returns a preview of the exact change and a confirmation token bound to those arguments; nothing is written. The assistant shows you the preview and only executes on your go-ahead with the token. On top of that, the key's scopes decide which write tools exist at all, and setting writes and live-traffic arming are separate per-key opt-ins.
No. Settings that hold credentials return only whether a value is set, never the value, and they cannot be written through MCP. To connect a marketplace account the assistant uses marketplace_connect_link_get, which returns a one time authorization link for the account owner to click, so no credentials pass through the assistant.
No. The MCP endpoint is on your own instance's subdomain and speaks directly to your assistant's client. InfiPlex does not run a model in the middle. The data your assistant sees is governed by your assistant vendor's data policy and the scopes on the API key you gave it.
It is included with API access on every InfiPlex plan. There is no AI add-on, no per-call fee, and no seat charge. See pricing.
Yes. API keys are per instance and per scope, so a 3PL can issue a read-only key to a client for their own assistant, keep a full key for its own operations, and revoke either independently. The connect-link tool lets the 3PL's assistant onboard a client's Amazon, Walmart, Etsy, TikTok Shop, or Shopify account without the client ever logging into InfiPlex.
When InfiPlex routes an order, it writes a step-by-step event log per line: which warehouses were eligible, which zip rules were checked, how much inventory each warehouse had at that moment, how many units were taken from each, and where any shortfall went. order_warehouse_routing_explanation_get returns that log, so the assistant explains the decision from the record rather than inferring it. The geo routing guide covers the rules themselves.
Yes. The MCP server is a protocol layer over the same REST API and the same key infrastructure. Scripts keep using REST; people use their assistant. A key can serve both, or you can issue separate keys with different scopes for each.
Tools outside the key's scopes are not listed, so the assistant does not see them. For the two extra gates (setting writes and arming) the tool exists but the server refuses the write with a message explaining which permission is missing, which the assistant relays to you. Nothing partial is written.
Resources
Bring a question you ask your operations team every day. We will connect Claude or Codex to a live InfiPlex instance and answer it, tool by tool, on the call.
Open any API key in your admin, set MCP Enabled to Yes, and follow the instructions shown on the key. Questions: Contact us, email info@infiplex.com, or call 888‑770‑0857.
Claude is a trademark of Anthropic, PBC. ChatGPT and Codex are trademarks of OpenAI. Cursor and Windsurf are trademarks of their respective owners. Model Context Protocol is an open standard. The demo above replays representative sessions with sample data; tool names, arguments, and the confirmation flow are shown exactly as the server implements them.
InfiPlex provides eCommerce companies, warehouses, and 3PLs with a robust platform to manage their sales channels.
©2026. All Rights Reserved. Privacy Policy | Terms & License