Model Context Protocol
NERMO runs a hosted MCP server — the same procurement engine behind the portals, exposed as tools your AI can call. Connect Claude, ChatGPT, or your own agent to your NERMO account, and port procurement becomes something you ask for in plain language.
MCP connection
Authorize your AI assistant to reach NERMO
Sign in to continue
Claude wants to connect to your NERMO workspace.
Work email
Password
Forgot?Trouble signing in? Contact your fleet admin
How it connects
No SDK, no custom integration. If your AI speaks MCP — Claude, ChatGPT, an IDE agent, a standalone procurement bot — it can reach NERMO today.
Paste NERMO's MCP endpoint into your assistant's connector settings — the same URL for every client that speaks the protocol. The connect guide has the endpoint and ready-made recipes per assistant.
/mcpYour AI sends you to NERMO's own sign-in. You authenticate with your NERMO account; the assistant never sees your password.
oauth 2.1 · pkceThe tools your assistant can see are scoped to your NERMO permissions — product search, quotes, orders, tracking, narrowed to what your account is granted.
scoped toolsConnecting is not an account. The MCP connection signs into a NERMO account that already exists — it never creates one. Operators are verified and onboarded before the first quote flows; if your fleet isn't on the network yet, start with Request access.
What it feels like
Under every plain-language request, your assistant is making structured, typed calls — the same operations your team sees in the Client Portal.
{
"tool": "request_quote",
"arguments": {
"port": "SGSIN", // Singapore
"eta": "2026-07-21T06:00Z",
"items": [
{ "name": "Provisions, fresh", "qty": 308, "unit": "crew-days" },
// …37 more lines, matched to the catalog
],
"idempotencyKey": "1c9e…d4" // retries can't double-order
}
}Draft an RFQ, refine it, send it, compare quotes, place the order, track delivery — the whole procurement loop is callable, not just a search box.
Free-text items are matched against NERMO's canonical catalog — marine stores and provisions — so "jasmine rice, 25kg bags" becomes the product vendors actually quote.
If a product is short at your port, your AI can ask NERMO for curated substitutes — and put a workable alternative on the order instead of a blank line.
Guardrails
An AI that can spend your money has to be held to account. The MCP surface runs under the same authorization, audit, and safety rails as everything else on NERMO.
Every MCP session belongs to a verified operator. Your AI can't order on behalf of a company NERMO hasn't vetted — and neither can anyone else's.
Tools run under your NERMO permissions, down to which vessels a session may act for. An assistant connected by a captain can do what that captain can do — nothing more. And no assistant can accept a contract for you: agreeing to NERMO's terms or starting a vessel's subscription is done by a person, in the portal.
Every write tool carries an idempotency key, so a dropped connection and a retried call can never turn one order into two.
Every call an agent makes is recorded — reads included, from the moment it connects. Actions that move money or goods land in the audit trail and your portal, where your team reviews agent activity exactly like human activity.
Running a headless agent? Standing procurement agents that work without a human in the loop get their own service-account credentials and scopes — see agents & integrators.
For suppliers
The same endpoint serves both sides of the network. Connect an assistant — or a headless agent beside your ERP — to your vendor account, and the working day arrives as conversation: new RFQs triaged, quotes priced and submitted, your catalog kept current.
Triage the RFQ inbox, price the lines from your catalog, and submit — the same quoting loop as the Vendor Portal, driven in plain language. Your quote reaches the operator exactly as you submitted it.
Push price and stock updates straight from an ERP export — the bulk upsert is built for exactly that — so the prices vessels compare are the prices you would actually charge.
Vendor sessions run under vendor permissions with the full guardrail set — scoped tools, idempotent writes, every call recorded. An agent quoting for you is held to account like an agent buying from you.
Your AI, your choice
Vessel names, IMO numbers, routes, and purchasing history are commercially sensitive — and who your AI answers to is part of that equation. MCP is an open protocol, and NERMO doesn't require any particular AI company to be in the loop.
Claude and ChatGPT work out of the box — but so does anything else that speaks MCP. The endpoint is the same for every client, whichever model is on the other end.
Run a model on your own infrastructure and the conversation with your AI never leaves it. Only the structured tool calls — an RFQ, an order, a status check — travel to NERMO.
OAuth sign-in, scoped permissions, the full audit trail — identical whether the client is a household-name assistant or a model behind your firewall.
NERMO × MCP
Get your fleet verified, connect the AI you already use, and start ordering by asking.
Whatever assistant your team adopts next, NERMO is one connector away.