More from Grafto
Linnworks MCP (source) by Grafto
Source · £249The exact TypeScript Cloudflare Workers codebase we run in production to connect Linnworks to Claude and other AI clients via the Model Context Protocol. Deploy it to your own Cloudflare account, customise it freely, and keep it — a one-off £249 with twelve months of free updates.
mcp-g.com · £249 one-off · For dev shops, agencies & in-house teams
What it does
No fluff — here is exactly what you get.
Every route, the auth layer and the full Linnworks API client. Yours to read, modify and deploy.
This is the code running against Green & Home Ltd’s own warehouse, not a demo.
Runs on Workers — fast at the edge and free for typical volumes on Cloudflare’s free plan.
Orders, stock, purchase orders, returns and customers — thirteen read tools plus four guarded write tools.
Dev shops and consultants can deploy for as many clients as they like under a single-purchaser licence.
Everything we ship to our own connector arrives in your portal as a re-download.
Plain English
MCP — the Model Context Protocol — is an open standard for giving an AI assistant a safe, structured door into a system. Without one, asking Claude about your Linnworks data means exporting a CSV and pasting it in. With one, Claude queries Linnworks directly, live, and can follow up on its own answer.
This product is the server that sits between the two. It speaks MCP to Claude on one side and the Linnworks.net API on the other, handling authentication, session tokens, pagination and the Linnworks quirks that make the API awkward. It runs on Cloudflare Workers, so there is no server to maintain and no idle cost — it scales to zero when nobody is asking anything.
You are buying the source, not a subscription. It deploys to your Cloudflare account, holds your Linnworks credentials, and keeps working whatever happens to us.
In practice
These are ordinary questions — the point is that you ask them in a chat window instead of building a report.
Claude searches open orders, groups them by status and age, and tells you which have been sitting longest — then drills into any one of them for the full detail.
Processed orders over a date range, cross-referenced against current stock levels per location, without exporting anything.
Search by email, name or postcode, then pull their order history to see whether a complaint is a first-time problem or a pattern.
Pending refunds and open RMAs listed out, with the orders they attach to — the pile that quietly costs money when it is ignored.
Search supplier POs and fetch one in full, so you can answer a “when is it coming?” email without opening Linnworks.
With the write tools enabled, set retail prices or archive discontinued lines — the handful of jobs worth automating once you trust the read side.
The tools
Every tool is described to the model so it can pick the right one for a question. The write tools are separate and deliberate — you can strip them out before deploying if you want a read-only connector.
search_open_orders · search_processed_orders · get_order — search live and historical orders by date range or criteria, and fetch one in full.
search_stock_items · get_stock_levels · get_stock_items_by_ids — find items, check levels per location, batch-fetch by id.
list_customers · search_customers — look people up by email, name or postcode.
search_purchase_orders · get_purchase_order — supplier POs, searchable and fetchable in full.
search_returns_refunds · get_order_refunds · get_actionable_refunds — RMAs, refunds on an order, and the pending pile.
create_order · set_retail_price · archive_stock_items · unarchive_stock_items — the four that change data in Linnworks.
Getting it running
You need a Linnworks.net account, a login at developer.linnworks.com, a free Cloudflare account and Node 18+. The guide walks through each step with the exact commands.
Create a System Integration application in the Linnworks developer portal and install it to your account.
Application id, secret, token and your Linnworks server URL go in as Wrangler secrets — never in the code.
One wrangler deploy and it is live on your own Cloudflare account, on your own subdomain.
Point Claude Desktop, Cowork or claude.ai at the endpoint and start asking questions.
The full user guide and troubleshooting notes are public — read them before you buy, not after.
Honestly
It is a source bundle, not an app. If you want something that installs itself, this is the wrong product and we would rather say so now than refund you later.
Skip it if: nobody on your side is comfortable with a terminal and a Cloudflare dashboard; you need a hosted service with support attached; you want a UI rather than a chat interface; or you are not already running Linnworks.
It fits if: you are an agency or dev shop serving Linnworks clients (one purchase covers unlimited client deployments); you are an in-house team that wants the connector inside your own infrastructure and your own data boundary; or you want a working, production-tested MCP implementation to learn from and extend.
Questions
No. The server runs on your Cloudflare account with your credentials. Data goes between your Linnworks account and your AI client. We never see it.
The complete TypeScript source, a one-off payment, and twelve months of updates by re-download. No subscription, no per-seat fee, no expiry — the code keeps working after the twelve months, you just stop getting new versions.
Yes — unlimited client deployments on a single purchase. That is the main reason agencies buy it.
Cloudflare Workers has a generous free tier and the server scales to zero, so typical volumes cost nothing. Heavy use moves you onto Cloudflare’s paid plan — check their current pricing.
Anything that speaks MCP — Claude Desktop, Cowork and claude.ai among them. It is an open standard, not a Claude-only integration.
Yes. The four write tools are separate; remove them before deploying and the connector can only read.
Drop us a line and we will let you know the moment it is ready.
