Connect Lexware Office to Claude: MCP for Lexoffice

What is an MCP server for Lexware Office (Lexoffice)?

An MCP server for Lexware Office, formerly Lexoffice, connects an MCP-capable AI assistant such as Claude to your Lexoffice instance through the open Model Context Protocol. The model can then access the Lexoffice Public API in a standardised way — read receipts, create invoices and manage contacts — without writing custom scripts or building Zapier workarounds for each new AI integration.

Lexoffice is particularly well suited to MCP: the API is REST-based, cleanly versioned, documented and quickly connected with an API key. That considerably reduces the effort for an MCP server.

Typical users: agencies, consultants, trades businesses, freelancers with a small team. Anyone working in Lexoffice wants bookkeeping out of the way, not as their main occupation. This is exactly where MCP comes in.

Use cases: what can a Lexoffice MCP server do in concrete terms?

1. Receipt capture and pre-coding A receipt arrives by email or drop. The AI agent reads it via OCR, compares it with existing suppliers, proposes category and tax account, and after approval creates the receipt in Lexoffice.
2. Invoice workflows "Create an invoice for client X for the services discussed in the pitch." The agent builds an invoice in Lexoffice, takes over address, tax rate and payment terms, and you review once and send.
3. Dunning proposals and open-items list The agent checks open items, prioritises by due date and customer value, and writes reminder emails as drafts, from friendly to firm, staggered by dunning level.
4. Reporting and liquidity view "How much do I have in outstanding receivables right now?" "How high were my software costs in Q1?" The agent pulls the figures from Lexoffice without you having to click around in the tool.
5. Supplier/customer CRM sync New contacts from the CRM land automatically in Lexoffice (or vice versa). Master data maintenance without double entry, and the agent reports when an IBAN, VAT ID or address is missing.
6. Handover to the tax advisor Month-end close as a guided routine: receipts checked? Bank reconciliation done? Any conspicuous accounts? The agent generates a short checklist, so your tax advisor has clean data.

How the integration works

Architecture view for Lexoffice:

AI agent ↔ MCP server (your EU host) ↔ Lexware Office Public API (API key, REST/JSON)

Concrete technical steps:

• Access: The account owner creates an API key for the integration and limits its permissions to the required functions. The key stays on the MCP server, outside chat and browser code.
• Tool mapping: We define MCP tools such as create_voucher, list_open_invoices, get_contact, create_invoice_draft. Each with a JSON Schema, validated inputs and clear error messages.
• OCR pipeline: For receipt capture, we put an OCR stage in front (e.g. Mindee, Azure Document Intelligence in the EU). The result flows into MCP tools in structured form.
• Read vs. write separation: By default the agent only creates drafts. Publishing / posting requires human approval, configurable.
• Logging: Every API action is logged with timestamp, user and diff.

GDPR, hosting & security

According to Lexware, Lexware Office stores data in high-security data centres in the Frankfurt am Main area, a good starting point. To keep the whole system GDPR-compliant, we pay attention to the following with every MCP server for Lexoffice:

• EU hosting of the MCP server (Hetzner or the Azure region Germany West Central in Frankfurt), no US edge.
• API key encrypted, rotation and revocation strategy defined.
• DPA with the AI provider with EU data residency (Claude via AWS Bedrock EU, OpenAI Enterprise EU, Azure OpenAI in an EU region).
• Data minimisation: The agent sees only the fields it needs for the task, with no blanket approval.
• Role and tenant separation if several companies run through one Lexoffice tax advisor access.

What does it cost?

The clean API keeps the integration effort manageable with Lexoffice. In concrete terms:

• First use case clearly scoped, then gradual expansion: typically receipt capture or an invoice workflow as the start.
• 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 volume (receipts per month), desired tools and whether OCR is integrated.

Connect Claude to Lexware Office: prerequisites and steps

For the integration described here, we use the official Lexware Office Public API. According to Lexware, this requires version XL. The MCP server provides the additional bridge between Claude and this API.

1. Check access: Confirm you have version XL and create an API key in Lexware Office with the required permissions. Do not share the key in chat.
2. Set up the MCP server: Store the key on the server. Initially, enable only the tools your task needs, such as reading outstanding invoices.
3. Connect Claude: Connect an accessible remote MCP server through a custom connector. For a local server, a suitable desktop extension is an option. Check your Claude version and your organisation’s permissions.
4. Limit write actions: Start with read-only access. Then let the assistant prepare invoices as drafts only; review, finalisation and sending take place in Lexware Office.
5. Verify the result: Start testing with a separate test account. Compare the displayed amounts and generated drafts directly with Lexware Office before connecting production data.

Setup depends on how you access Claude. Anthropic explains desktop and web connectors separately. For other AI assistants, we also check MCP support, transport and authentication; switching models alone does not establish a connection.

What the Public API can do with invoices — and what remains separate

The Lexware API documentation describes creating invoices as drafts or as finalised invoices immediately. However, an existing draft cannot subsequently be finalised through the API. Creating an invoice does not automatically send an email either. For our initial setup, approval, finalisation and sending therefore remain in Lexware Office.

Common pitfalls from practice

The basic integration is quickly built. It only becomes stable in everyday use once these points are solved:

• Rate limits: The Lexoffice API throttles requests. Bulk operations, such as catching up on 200 receipts, need a queue with retry logic, otherwise runs break off midway.
• Duplicates: Without an idempotency check, the agent creates the same receipt twice if a response gets lost. Before every creation, we therefore check against receipt number and amount.
• Invoice status: An API draft cannot subsequently be finalised through the Public API. Plan to finalise it in Lexware Office; creating it again must not produce a duplicate invoice.
• OCR confidence: Recognised amounts and tax rates are run against plausibility rules (totals, VAT rates, known suppliers) before they become a posting proposal.
• Token management: Expired or revoked credentials have to be actively detected and reported, otherwise actions fail silently and receipts are left waiting.

These are exactly the points that distinguish a weekend script from a system you can trust with your bookkeeping.

Frequently asked questions

Which Lexoffice API does the MCP server use?

We use the official Lexware Office Public API (REST/JSON, access via API key). According to Lexware, it is available from the XL plan; whether your plan fits, we clarify in the audit.

Can the agent send invoices directly?

Creating an invoice through the Public API does not send an email. We initially let the assistant prepare a draft; you review, finalise and send it in Lexware Office. Additional automated sending would need a separate integration and approval.

How is OCR handled?

For receipt capture we integrate an OCR engine with EU data residency. The result flows into MCP tools in structured form: the agent does not see the image but processed fields.

Does this also make sense for tax advisors with Lexoffice clients?

Yes. Through the tax advisor access, several client accounts can be connected separately, with clear permission boundaries per client.

Is Lexoffice + MCP GDPR-compliant?

Yes, if the whole chain is right: according to the vendor, Lexware Office stores data in data centres in the Frankfurt am Main area, the MCP server runs in the EU, a DPA with EU data residency is in place with the AI provider, and the agent sees only the fields it needs for the task. Whether a US provider is in the path depends on the chosen model access; we determine that in advance.

How quickly is a Lexoffice MCP integration productive?

We start with a clearly scoped use case. We set the timeline after the AI Check. The clean API keeps the integration effort manageable.

Which AI models work?

What matters is the AI assistant or application: it must support MCP as well as the server’s transport and authentication. Claude offers desktop and remote connectors. For other applications, we check their specific support before connecting them.

Can I connect Claude Desktop directly to Lexoffice?

Claude needs an MCP server to bridge the Lexware Office Public API. Depending on how you run it, you connect it as a remote connector or through a suitable local desktop extension. API access and enabled tools are also required. The section “Connect Claude to Lexware Office” above explains the process.

Are there ready-made Lexoffice MCP servers to try out?

There are community implementations of very varying maturity. For business use, auth, logging, approvals and GDPR hardening are what count, and they make up most of the work. What a robust server has to be able to do is set out in the section "Common pitfalls" above.