Integration · Shopware

Connect Claude to your Shopware shop.

The deepest Shopware toolset an LLM can hold: storefront and admin reads, CMS and landing-page authoring, and the full order state machine. Claude proposes; nothing changes on the shop until you sign it off — and every change lands in the ledger.

Start freeRead the guide

Claude Desktop · Claude Code · ChatGPT connectors · Gemini CLI · any MCP client

Claude Desktop · mcp.datavessel.io
> which orders are stuck in payment for more than 5 days?
sw_admin_list_orders … 4 orders, oldest 2026-08-14
> cancel #10412 and #10437, restock both
sw_admin_transition_order_state → waiting for your sign-off
✓ approved from the inbox · 2 rows written to the ledger
The tool catalogue

37 typed tools. 13 of them change the store — every one behind sign-off.

Storefront and Admin API as one typed catalogue. Reads run free; the 13 writes go through the sign-off inbox, and state changes follow Shopware's own state machine — the legal transitions are read first, then taken.

01

Read the shop

  • Orders, products, customers, stock
  • Storefront: navigation, search, product listings
  • CMS pages and landing pages
  • Legal state transitions per order
02

Run the orders

  • Ship deliveries with tracking codes
  • Cancel, complete, reopen via state transitions
  • Record returns and refunds through the payment state machine
03

Author the content

  • Create products — inactive by default
  • Update product copy, meta titles, keywords
  • Create and update CMS pages and landing pages
Three ways in

The same tools, wherever the work happens.

One connection; every surface reads and writes through the same typed catalogue, with the same gates.

From Claude, over MCP

Claude Desktop, Claude Code or any MCP client connects to one endpoint and sees the Shopware tools for the shop you scope it to.

The MCP endpoint

From the terminal

The datavessel CLI runs the same catalogue as commands and ships a Claude Code plugin whose read agents are structurally blocked from writing.

The CLI

In the operations center

Policies subscribe agents to shop events and schedules — Order Desk, product-copy refresh, SEO audit — asking you first, per store.

The use cases
Why not just an API key

An LLM with your admin credentials is a liability. This is an operations layer.

01

Writes wait for you

A raw Admin API key lets a model do anything, silently. Here every write pauses in the sign-off inbox — approve from the web or the mobile app — until you trust a job enough to put it on autopilot.

02

Scoped per shop

Tools are scoped to one store at execution time. An agency running several Shopware shops keeps every client's data and credentials isolated — one shop's run cannot touch another's.

03

Every change in the ledger

What changed, on which shop, approved by whom, for how much. Reads aren't listed — the ledger is a record of changes — and it compiles into the per-client report.

Already built on these tools

The jobs, ready to deploy.

Each ships as an agent with the policy that runs it — asking you first, per store, with every action in that client's ledger.

Questions

Asked before connecting.

How do I connect Claude to Shopware?

Connect the shop to datavessel once (Admin API credentials, a few minutes), then point Claude Desktop, Claude Code or any MCP client at the datavessel MCP endpoint. Claude sees the Shopware tools for that shop immediately — no plugin to install on the shop itself.

Can Claude change my shop?

Only through the 13 write tools, and every write waits in the sign-off inbox until a human approves it. Order-state changes follow Shopware's own state machine, new products are created inactive, and each approved change is recorded in the ledger.

Which Shopware versions are supported?

Shopware 6, via the Store API and Admin API. Both self-hosted and cloud shops work — the connection is API credentials, not a shop plugin.

Does it work with ChatGPT or Gemini, not just Claude?

Yes. The endpoint is standard MCP, so any MCP client works — Claude Desktop and Claude Code, ChatGPT via connectors, the Gemini CLI, Cursor and others. Inside the operations center, agents run on whichever key you bring: Anthropic, OpenAI or Google.

What does it cost to run?

You bring your own AI key (Anthropic, OpenAI or Google) and pay the model provider directly — datavessel adds no markup on tokens. Plans are priced by connected stores, free for one store.

Is my shop data used for training?

Your data is isolated per customer and scoped per store, and model calls run on your own key under your provider's terms. See the security page for the full commitments.

Your Shopware shop, one sign-off away.

Connect the shop, ask Claude for the first sweep, and approve what goes out — every change on the record.

Start freeAll integrations

Free for one store · 14-day trial on paid plans · no card · bring your own AI key