.svg)
.webp)
Connect Shopware to ChatGPT, Claude and Grok: The Practical Guide 2026
Your customers research products with ChatGPT, Claude or Grok. Your team can use the same assistants to work with your own shop data: checking stock, finding missing product descriptions, adjusting prices. Shopware now ships its own interface for this, right in the core. In this guide, we show how the connection works today, where the limits are, and what to watch out for in terms of security and data protection.
All information reflects the state of October 2026 and links to the official sources.
- Shopware ships its own MCP server in the core: introduced in 6.7.11, always enabled from version 6.7.14.0.
- Two endpoints: /api/_mcp for the back office and /store-api/_mcp for the shopper view.
- Claude is the easiest to connect. ChatGPT usually needs a layer in between because of OAuth. Grok works via connectors or the xAI API.
- Security first: a dedicated ACL role, the MCP allowlist, read-only tools to start with and a test on staging.
- Buying directly in ChatGPT isn't possible at the moment. Agentic Commerce and UCP still prepare your shop for AI shopping.
Why connect your Shopware shop to AI assistants?
An AI connection gives you two quite different benefits.
1. Internal: work faster with real shop data. Instead of exporting data and filtering spreadsheets, you ask your question directly in the chat. The Shopware documentation itself gives examples such as:
- "Find products with stock below five units."
- "Find all active products without descriptions."
- "Generate SEO-friendly meta titles for the products in a category."
- "Increase prices of products in a specific category by 5%. Show a preview before applying the changes."
2. External: become discoverable in AI shopping. AI assistants are increasingly used for product search. With the "Agentic Commerce" extension, Shopware provides product feeds and an open protocol so that AI platforms and AI agents can find and process your catalogue.
On top of that, Shopware Copilot works directly inside the Administration. We cover all three routes below.
How it works: Shopware's MCP server
What is MCP?
The Model Context Protocol (MCP) is an open standard that lets AI applications work with external systems. Instead of copying data out of a chat window, an assistant calls defined tools, such as a product search, and receives structured results. ChatGPT, Claude and Grok all support MCP.
Shopware's native MCP server
Shopware ships a native MCP server in its core. It was introduced in version 6.7.11.0 behind the MCP_SERVER feature flag. From version 6.7.14.0, the flag is gone and the server is always enabled. That release also added "progressive tool discovery": an AI client initially sees only a few discovery tools and enables the toolsets it needs on demand.
Two endpoints: Admin API and Store API
- /api/_mcp – back-office access via the Admin API: products, orders, settings and media. Permissions follow Shopware's access control lists (ACL).
- /store-api/_mcp – access from a shopper's perspective in a sales channel. According to the Shopware documentation, it's meant for agents that browse a shop and work with the same data the storefront shows.
Integrations, roles and allowlists
Authentication works through a Shopware integration. You get an access key and a secret, which the AI client sends as HTTP headers (sw-access-key and sw-secret-access-key). These credentials don't expire as long as the integration exists. Admin API bearer tokens also work, but they expire after ten minutes by default.
Three points matter most for security:
- A dedicated role instead of full access. Shopware's docs show the setup with admin rights, but explicitly recommend a dedicated ACL role with only the necessary permissions for production use.
- MCP allowlist. Under Settings → Integrations, you define per integration which tools, resources and prompts the AI may use at all. Individual users have their own allowlist.
- Rate limiting. According to the configuration docs, both endpoints have their own limits to protect against brute-force attempts and runaway agent loops.
What "experimental" means
Whichever AI assistant you connect, the setup always follows the same process:
Step by step: connecting Claude to Shopware
Claude currently offers the most direct route. The reason: its "custom connectors" let you store fixed credentials as request headers, which is exactly how Shopware's MCP server authenticates. According to Anthropic, custom connectors are available on the Free (one connector), Pro, Max, Team and Enterprise plans.
On a Pro or Max plan:
- In Shopware, create an integration with a restricted role and set the MCP allowlist.
- In Claude, open Customize → Connectors, click "+ Add", then "Add custom connector".
- Enter a name and your shop's URL, for example https://your-shop.com/api/_mcp.
- Review the detected authentication settings and choose "No sign in" if needed.
- Under "Request headers", add the values for sw-access-key and sw-secret-access-key.
- Click "Add". In a chat, enable the connector via "+" → Connectors.
On Team and Enterprise plans, owners add the connector under Organization settings → Connectors. Members then connect individually.
- Claude connects from Anthropic's cloud, not from your computer. Your shop must therefore be publicly reachable. Behind a firewall, you need to allowlist Anthropic's IP addresses.
- When a team shares a fixed key, every user has the same permissions. Keep the integration's role especially narrow.
For developers, the Shopware docs also describe local setup in Claude Desktop (via the claude_desktop_config.json file) and in Claude Code.
ChatGPT and Shopware: two routes and an honest caveat
With ChatGPT, it's worth separating two goals: working with shop data, and making products visible in ChatGPT.
Route 1: Work with shop data in ChatGPT (MCP)
According to OpenAI's developer docs, you add a custom MCP server like this:
- Open chatgpt.com/plugins, click the plus button, then "Add custom MCP server".
- Enter a name and description, and the server's public HTTPS URL, or choose a Secure MCP Tunnel.
- Configure authentication, confirm the risk warning and select "Create as a plugin".
What's available depends on your plan and your workspace policies. In Business and Enterprise workspaces, admins enable this capability.
For developer teams, there's a shortcut: OpenAI's coding agent Codex is documented in Shopware's docs with a ready-made configuration. Via the OpenAI API, MCP servers can also be used as tools in your own applications.
Route 2: Make your products visible in ChatGPT (Agentic Commerce)
With Shopware's free "Agentic Commerce" extension, you create a dedicated sales channel from the OpenAI template. There, you map properties such as colour, size or material, add a mandatory URL for your return policy, and generate a JSONL product feed. You submit the feed URL at chatgpt.com/merchants. It must be publicly accessible without authentication.
- According to Shopware, registration for the ChatGPT marketplace is currently only available in the US and not in Europe.
- OpenAI ended native "Instant Checkout" in ChatGPT in March 2026, as CNBC among others reported. Since then, purchases happen in the merchant's own shop.
Preparing the feed is still worthwhile. You need clean product data and property mappings anyway, and the same extension also provides a Google product feed.
Connecting Grok to Shopware
In the Grok app
According to the xAI documentation, connectors are available to all Grok users. You add your own MCP server like this:
- Open grok.com/connectors.
- Click "New Connector" and select "Custom".
- Enter the MCP server URL and complete any required authentication.
Grok then discovers the available tools automatically. On Grok Business and Enterprise, a team admin must first provision the connector in the cloud console. Here, too, the server must be publicly reachable.
Via the xAI API
For your own automations, the xAI API offers "Remote MCP Tools". You specify the server URL and can send additional headers, including your Shopware credentials. With the allowed_tools parameter, you expose only specific tools. That matters because the xAI API doesn't currently support the require_approval parameter, so there's no built-in confirmation step.
ChatGPT, Claude and Grok compared
All three assistants speak MCP. The main difference is authentication, and with it the setup effort:
| Claude | ChatGPT | Grok | |
|---|---|---|---|
| Setup | Customize → Connectors → "Add custom connector" | chatgpt.com/plugins → "Add custom MCP server" | grok.com/connectors → "New Connector" → "Custom", or the xAI API |
| Authentication | Fixed credentials as request headers | OAuth 2.1 per the MCP specification | App: depends on the server; API: additional headers possible |
| Shopware keys usable directly? | Yes | Usually not – needs an OAuth layer | Yes, via the xAI API |
| Availability | Free (1 connector), Pro, Max, Team, Enterprise | Depends on plan and workspace; Business/Enterprise: enabled by admins | Connectors for all users; Business/Enterprise: admin provisions |
| Effort | Low | Medium (development) | Low to medium |
Shopware Copilot and the Agentic Commerce extension
Shopware Copilot: AI right inside the Administration
If you don't want to connect an external assistant, you can use Shopware Copilot. It answers questions about Shopware features on all plans. With "Data Insights", it delivers numbers and charts on customers, orders and shop performance. With "Copilot Agentic", it carries out tasks such as discount campaigns, flows or bulk product updates. Changes are prepared as drafts and only applied once you confirm them.
With Shopware 6.7.15.0, Copilot Agentic moved from beta to stable, according to the October 2026 release news. Copilot can now be assigned to specific ACL roles, and apps and plugins can make their capabilities available to Copilot via MCP. The advanced features require a Shopware Intelligence+ subscription.
Agentic Commerce: product feeds and UCP
The "Agentic Commerce" extension (beta) has two building blocks:
- Product feed sales channel: feeds for OpenAI and Google, with tracking of orders, new customers and revenue from AI channels via affiliate and campaign codes.
- Universal Commerce Protocol (UCP): an open protocol that lets AI agents discover your shop, browse the catalogue and buy on a shopper's behalf. You enable it per sales channel and define what agents may do: browse the catalogue, use the cart, apply discounts, check out and read order status. Available transports are REST, A2A, Embedded and MCP.
AI chatbots from the Shopware Store
If you mainly want to use AI for your customers in the storefront, a chatbot plugin is often the faster route. Two examples from the Shopware Store:
- All-in-One AI Shop Assistant by Biloba IT: customer guidance in the storefront, product and SEO copy, and image alt texts. You can choose between ChatGPT, Gemini, Claude and Mistral, and it also offers MCP integration and AI labelling in the chat.
- NextGen OpenAI Chatbot und Support Agent by 5 Elements Websolutions: 24/7 support and natural-language product search based on live shop data, managed via a free Agents Manager in the back end.
When choosing, check compatibility with your Shopware version, consent and deletion features, and the AI provider's ongoing API costs.
Checklist: security and data protection
Frequently asked questions (FAQ)
Which Shopware version do I need?
The MCP server is available from Shopware 6.7.11.0 (with a feature flag). We recommend at least 6.7.14.0, where it's enabled by default and supports progressive tool discovery.
Is the MCP server suitable for a live shop?
It's labelled experimental until Shopware 6.8. With a narrowly scoped role, an allowlist, read-only tools to start with and prior testing on staging, the risk can be kept well under control.
Which AI assistant is easiest to connect?
Currently Claude, because your Shopware credentials can be stored directly as request headers in a custom connector. With ChatGPT, the OAuth requirement usually calls for a layer in between.
Can customers buy from me directly in ChatGPT?
Not at the moment. OpenAI ended Instant Checkout in March 2026, and according to Shopware, merchant registration for the ChatGPT marketplace is currently only available in the US. With UCP, however, Shopware already offers an open protocol through which compatible AI agents can complete a checkout.
Does my shop need to be publicly reachable?
For the cloud versions of Claude, ChatGPT and Grok, yes, since they connect from their own servers. For private servers, OpenAI offers Secure MCP Tunnel, and xAI documents tunnelling options. For Claude, you may need to allowlist Anthropic's IP addresses.
What does an AI connection cost?
That depends on the route: your plan with the AI provider, possibly Shopware Intelligence+ or plugin subscriptions, and the setup effort. You'll get a concrete estimate for your shop in a free initial consultation.
Request an individual consultation
Every shop is different. Which Shopware version you run, what data the AI may see, and whether you want to become more efficient internally or more visible externally all determine the right route.
As a Shopware Bronze Partner based in Munich, Deepware supports merchants across Germany, Austria and Switzerland with exactly these questions:
Tell us briefly about your situation: click “Contact us” and fill in the short form. We'll get back to you personally. The initial consultation is free and non-binding.
Revolutionizing Large Language Models with Mixture-of-Experts Architecture
In the rapidly evolving landscape of artificial intelligence, Tencent has unveiled a game-changing innovation: Hunyuan A13B. This open-source large language model represents a paradigm shift in how we approach AI efficiency, combining the power of 80 billion parameters with the computational efficiency of just 13 billion active parameters through its revolutionary Mixture-of-Experts (MoE) architecture.
Technical Specifications
The model employs a sophisticated fine-grained MoE architecture with one shared expert and 64 non-shared experts, activating 8 experts per forward pass. It features 32 layers, SwiGLU activations, and Grouped Query Attention (GQA) for efficient memory utilization.
Unique Selling Propositions
Performance Comparison
Benchmark Performance Visualization
Key Use Cases
Competitive Advantages
Efficiency Comparison
(Performance per Billion Parameters)
Future Implications
Hunyuan A13B represents a significant step forward in democratizing AI technology. Its efficient architecture and open-source nature are likely to:
Hunyuan A13B stands as a testament to the power of innovative architecture in AI development. By combining the efficiency of Mixture-of-Experts with dual-mode reasoning and a massive context window, Tencent has created a model that challenges the conventional wisdom that bigger always means better.
For organizations looking to implement advanced AI capabilities without the computational overhead of traditional large language models, Hunyuan A13B offers a compelling solution. Its open-source nature, combined with state-of-the-art performance, positions it as a game-changer in the AI landscape.
Hunyuan A13B is available now on Hugging Face and can be deployed using popular frameworks like Transformers. Join the growing community of developers leveraging this powerful model for innovative AI applications.
Transforming Text into Cinematic Reality with Native Audio Integration
The Next Frontier of AI Video Generation
In May 2025, Google DeepMind unveiled Veo 3, a groundbreaking AI video generation model that has fundamentally changed how we think about artificial content creation. This state-of-the-art system doesn't just generate videos—it creates complete audiovisual experiences that blur the line between AI-generated content and reality.
Key Statistics & Performance Metrics
AI Video Generation Market Evolution
Veo Model Comparison: Key Capabilities
Revolutionary Features
Performance Benchmarks: Veo 3 vs Competitors
Technical Specifications
Development Timeline
Google AI Plan Comparison
Plan Details
User Adoption Rate (First Month)
The response to Veo 3 has been unprecedented in the AI video generation space. Within just three weeks of launch, the tool has:
- Generated over 1 million videos across all user tiers
- Achieved 85% user satisfaction in early beta testing
- Reduced video production costs by 70% for small content creators
- Sparked industry-wide discussions about AI authenticity and regulation
Challenges and Limitations
Future Implications & Industry Impact
Veo 3 represents more than just a technological advancement—it signals a fundamental shift in content creation. The integration of native audio generation sets a new industry standard that competitors will struggle to match.
Predicted Industry Changes:
- Content Creation Democratization: High-quality video production becomes accessible to non-professionals
- Traditional Media Disruption: Lower barriers to entry challenge established production companies
- Regulatory Response: Governments likely to introduce stricter AI content labeling requirements
- Educational Revolution: Personalized video content transforms online learning
- Marketing Evolution: Brands can create unlimited variations of video advertisements
Conclusion
While challenges remain around cost, accessibility, and ethical implications, Veo 3 has undeniably set the new standard for AI video generation. As we move forward, the question isn't whether AI will transform video content creation—it's how quickly the industry will adapt to this new reality.
The future of video content creation is here, and it's more accessible, more powerful, and more realistic than ever before.
Most online stores have streamlined their ordering process to just one click. For years, however, the process of canceling an order has gone in the opposite direction: finding a form, typing an email, and waiting for a reply. The legislature is putting an end to this very imbalance—and it will happen soon.
With Directive (EU) 2023/2673, the European Union is amending the Consumer Rights Directive and introducing a mandatory digital withdrawal function. In Germany, it will be implemented through the new Section 356a of the Civil Code (BGB), which will take effect on June 19, 2026—across the EU, uniformly, and without a further transition period. Austria is implementing the same requirements through the Consumer Rights Directive Implementation Act.
To many companies, this sounds like a minor front-end issue. In practice, however, it affects the entire process—from the user interface to customer accounts, CRM, and ERP, all the way through to refunds and audit-proof documentation. Tackling this early on helps avoid a last-minute project under time pressure.
Why bother:
A Response to “Dark Patterns”
In recent years, European regulators have consistently taken action against contractual practices that pressure people into signing agreements while making it unnecessarily difficult for them to exercise their rights. Making a purchase takes seconds—but canceling it has become a test of patience. The new rule levels the playing field. Legal experts call this functional equivalence: The process of getting out of a contract must not be more complicated than the process of entering into it.
Anyone who can enter into a contract online should also be able to cancel it online.
Important to know: The law refers to a “cancellation function,” not literally to a “button.” In practice, this can be a clearly labeled button—or a prominently displayed link, as long as it is just as easy to find. What matters is not the form, but the accessibility.
Who is affected:
Much more than a traditional online store
Many merchants assume that the requirement applies only to pure online stores. The scope is broader: it applies to distance contracts for goods, services, and financial products concluded via an online user interface—that is, via websites as well as apps. Specifically, this includes, among other things:
And it's not just about where a company is headquartered. Providers outside the EU—such as those in Switzerland—are also affected as soon as they specifically target consumers within the European Union.
What Needs to Be Done
Four Requirements, One Process
A link to a PDF or a contact page isn't enough. What's expected is a seamless digital process—simple, transparent, and easy to follow.
One point that is often overlooked: When the feature changes, the required text changes as well. The cancellation policy and privacy policy must be updated—the latter with regard to the newly collected data, its processing, and retention period.
What Many Misjudge
Platform Location
Native function or custom integration?
As the deadline approaches, apps, plugins, and extensions are popping up everywhere. The situation varies depending on the system:
A native platform feature lowers the barrier to entry. But as soon as ERP and CRM integration, automated returns and refunds, custom business logic, multilingual workflows, audit trails, or existing customer portals come into play, a standard app quickly reaches its limits. Larger stores and custom solutions require integrations that fit seamlessly into their processes.
Even though some details of the national implementation are still being worked out, the essentials are clear: The “Cancel” button will be mandatory. It is not an isolated exception, but part of a broader trend toward greater transparency, consumer-friendliness, and digital standards in European commerce. Those who modernize their processes now will not only ensure compliance—but also create a better experience for their own customers.
Here's how Deepware helps
Not just a button. A robust process.
The task is not to incorporate a specific element, but to establish a revocation process that complies with the law and is technically sound. We will prepare your systems for the new requirements in a timely manner.
.svg)
