What is an MCP server for JTL-Wawi?
JTL-Wawi is an ERP and inventory management system widely used among German e-commerce mid-sized companies, from the one-person shop workshop to the mail-order retailer with its own logistics centre. It combines items, stock, orders, customers, shipping and multichannel connection to shop, Amazon and eBay in one system. An MCP server for JTL-Wawi makes this data and these actions available to an AI agent via the open Model Context Protocol, in a standardised way, so you do not have to maintain a separate script for every model and every use case.
Important up front: JTL runs an official MCP server at https://developer.jtl-software.com/mcp which, according to JTL, only searches and reads the developer documentation (JTL: Build with AI). It helps with programming against the API but does not access your items, stock or orders. For data access there is the JTL-Wawi REST API: locally at http://127.0.0.1:64110/api/eazybusiness/ with an API key from a two-step registration, in the cloud at https://api.jtl-cloud.com/erp/ with OAuth 2.0 client credentials (JTL developer documentation). We build the MCP layer on top of that for each project. So JTL-Wawi is not "MCP-ready out of the box", but it can be cleanly connected via the official API.
Typical users: online retailers, multichannel shippers, manufacturers with their own D2C shop. Anyone running JTL-Wawi wants to get rid of routine clicks (stock queries, order status, reporting) without leaving the familiar system.
In one sentence: JTL-Wawi + MCP means: you ask the agent "Which items are below minimum stock and have open orders?" and it pulls the answer live from the Wawi API, instead of you building a report.
Use cases: what can a JTL MCP server do in concrete terms?
- Stock and availability queries "How many units of SKU 4711 are spread across all warehouses?" The agent reads stock from the Wawi, takes reserved quantities from open orders into account and reports what is really available, without a warehouse management screen.
- Order monitoring and escalation "Which orders from yesterday are still in 'in progress' status?" The agent filters orders by status, platform and date, detects shipping delays and proposes a prioritisation, especially valuable in peak periods.
- Item maintenance and master data Create new items or enrich existing ones: the agent pre-fills description, category, dimensions and attributes, checks for duplicates by item number and creates the item after approval. Master data quality without copy-paste.
- Customer and order history "What has customer Mustermann ordered in the last twelve months, and are there any open complaints?" The agent pulls customer data and linked orders and delivers a summary for support, instead of a manual search.
- Reorder and planning proposals The agent compares sales velocity, current stock and minimum stock, identifies items close to selling out and creates a purchase proposal per supplier, as a draft that purchasing only has to review.
- Reporting in plain language "How has revenue for category X developed compared with the previous month?" The agent aggregates order data and answers in sentences rather than a pivot table: a quick liquidity and performance view in between tasks.
How the integration works
Architecture view for JTL-Wawi:
AI agent ↔ MCP server (your EU host or your network) ↔ JTL-Wawi REST API (local or via the JTL cloud API) ↔ JTL-Wawi
Concrete technical steps:
- Setting up API access: Locally, we register the application in the two-step process and receive an API key; according to the JTL documentation, the endpoint listens on
127.0.0.1:64110, so the MCP server runs on the same machine or is connected via VPN or a secured reverse proxy. In the cloud we use OAuth 2.0 client credentials. - Tool mapping: We define MCP tools such as
get_stock,list_orders,get_customer,create_item_draft, each with a JSON Schema, validated inputs and clear error messages, so the agent does not reach into the void. - Mapping to the JTL data model: JTL cleanly separates item, stock, order and customer. Where the REST API does not cover a field, we selectively add the classic JTL connector or a read-only SQL view.
- Read vs. write separation: By default the agent delivers analyses and drafts. Write actions relevant to bookkeeping, such as stock corrections or order status changes, require explicit approval, configurable per tool.
- Logging: Every API action is logged with timestamp, user and diff, important because stock and order data are directly business-critical.
Note on API coverage: JTL keeps extending the REST API. In the audit we check for each use case whether the required object is already covered by the REST API or whether the proven JTL connector needs to be added. That way you avoid functional gaps after go-live.
GDPR, hosting & security
In most businesses JTL-Wawi runs on-premises or on EU hosting, so customer, order and stock data are in your hands anyway. To make sure the AI extension does not undermine this, we pay attention to the following with every MCP server for JTL-Wawi:
- EU hosting of the MCP server (Hetzner or the Azure region Germany West Central in Frankfurt) or operation directly next to the Wawi in your own network, with no US edge.
- Secured API access: no openly reachable API port, but local operation, VPN or a reverse proxy with authentication and IP whitelisting.
- DPA with the AI provider with EU data residency: Claude via AWS Bedrock (EU region Frankfurt), Google Vertex AI (EU), OpenAI Enterprise EU or Azure OpenAI in an EU region.
- Data minimisation: The agent sees only the fields the respective use case needs; stock queries do not need customer data in clear text.
- Tenant and role separation if several JTL tenants or warehouses run through one instance.
What does it cost?
Because JTL-Wawi does not come with a ready-made MCP server, the share of API connection and mapping work is somewhat larger here than with pure cloud tools, but the system then stays very robust once built. In concrete terms:
- First use case clearly scoped, then gradual expansion: typically stock queries or order monitoring as a starting point, because they are read-only and quick to take effect.
- Clear scope per use case, no hours roulette. Scaling after that in small, clearly defined steps.
We give a concrete figure in the initial consultation, depending on order volume, number of warehouses, desired tools and whether the JTL connector is integrated alongside the REST API.
Related guides
Frequently asked questions
Is there a ready-made MCP server for JTL-Wawi?
Not for your Wawi data. JTL runs an official MCP server at https://developer.jtl-software.com/mcp which, according to JTL, only searches and reads the developer documentation; it does not access items, stock or orders. For data access we connect JTL-Wawi via the official API and build the MCP layer on top of it for each project.
Which JTL-Wawi interface is used?
The REST API of JTL-Wawi for items, stock, orders and customers. According to the JTL developer documentation, it runs locally at http://127.0.0.1:64110/api/eazybusiness/ with an API key from a two-step registration; for the cloud there is https://api.jtl-cloud.com/erp/ with OAuth 2.0 client credentials. Where the API does not cover an object, we add the classic JTL connector.
Can the agent change orders or stock directly?
Technically that is possible. In practice, for business-critical write actions such as stock corrections and status changes, we recommend a short approval step. Read queries and proposals run fully automatically, and posting stays under control.
Does this also work with on-premises JTL installations?
Yes, that is actually the rule. Since JTL-Wawi often runs locally or on a dedicated server, we secure API access through local operation, VPN or an authenticated reverse proxy. The API port is never openly reachable on the network.
Can JTL-Wawi + MCP be operated in a privacy-compliant way?
Yes, if the chain is right: JTL-Wawi usually sits in your own infrastructure, the MCP server runs in the EU or in your own network, the AI provider is integrated with EU data residency and a DPA, and the agent sees only the necessary fields. We document where each part of the chain processes data before the start.
How quickly is a JTL MCP integration productive?
The timeline depends on data volume, desired tools and write permissions. We set it after the AI Check. Read-only use cases come first, followed by write workflows with approval logic.
Which AI models work?
All MCP-compatible models: Anthropic Claude, OpenAI GPT, Google Gemini, Microsoft Copilot. Switching models is possible without a new integration.
Want an MCP server for JTL-Wawi?
30-minute initial consultation: we check whether your JTL instance is API-ready, which use cases carry first, and what the effort realistically looks like.
Book an AI Check arrow_forward