Is there an MCP server for weclapp?

MCP stands for Model Context Protocol, an open standard that lets AI applications access external systems. Anthropic introduced it and handed it over to the Agentic AI Foundation of the Linux Foundation on December 9, 2025. An MCP server is the counterpart: it defines which tools an AI client such as Claude or ChatGPT may call in your weclapp, i.e. whether the AI only reads or also writes.

On October 12, 2026, we checked the 2026 changelog, the API changelog, the search in the weclapp Store, and the search for "MCP" on weclapp.com and in the support portal. We found no entry for an MCP server. That is our check result, not a statement from weclapp, and we did not evaluate the roadmap.

There are also community projects. On GitHub, we found six hits for the search "weclapp mcp" on October 12, 2026, plus directory entries. You run most of these projects yourself; there are also hosted services. Some only read, others also change data, and how a change is confirmed is handled differently by each project. Three questions help with any solution:

• Who runs the server, and who keeps it up to date?
• Does the AI only read, or does it also write? How do you confirm a change?
• With whose permissions and under what name does it act in weclapp?

wals.pro AI 4 weclapp is a hosted MCP server for your weclapp. wals.pro is a weclapp partner. Setup has three steps:

1. Register. We review every registration personally and then enable your access. Then, in the dashboard, you store the address of your weclapp tenant and an API key. The key is kept in a secret store and is never shown to you again.
2. Enter the server address in your AI client. It is the same for all clients and is shown in the dashboard.
3. Sign in via OAuth. The AI client then connects to your tenant.

If your weclapp restricts access by IP address, we will give you our server's fixed outbound addresses on request via Help. The Setup Center guides you up to the first query (AI when setting up weclapp); more about the product is on the wals.pro AI 4 weclapp product page.

Claude, ChatGPT and Microsoft Copilot (via Copilot Studio) connect to remote servers; Claude reaches them from the Anthropic cloud, so the server must be publicly reachable. You can also connect developer clients such as Claude Code, Cursor or VS Code with the same server address; the instructions are in the dashboard. The provider's documentation is authoritative.

A fictional example, derived from the tool description; the flow may differ depending on the AI client and model. The sales manager of a trading company asks in Claude:

1. The AI queries quotes with status "open" and a quote date more than 30 days ago and fetches the customer names as well.
2. You get a list in the chat.
3. It is good if the AI states how it read "for 30 days": as quote date or as validity date. If it doesn't, ask and check the interpretation.

That was read-only: no preview, no approval, nothing changes in weclapp. More examples: Questions about contracts and tickets, ERP for IT system integrators.

If you want to create a follow-up task for a quote, the AI first shows a preview. You confirm it in the chat, and only then should the AI trigger execution. The server executes exactly this preview. The approval is valid once and only for a short time. It is bound to the content via SHA-256 (digital fingerprint): if a value changes, the server rejects it. This applies to creating, changing and deleting.

The server checks whether the approval matches the preview. Whether you are asked before execution is controlled by your AI client. Claude asks, but offers "Allow always". Don't enable that for write actions, and read every preview before you confirm.

The log is written on every call (who, when, which action, which result), without the contents of your records. How much of it you can view and export in the dashboard depends on your wals.pro plan.

Some actions can automatically trigger emails in weclapp, such as re-confirming an order. For this, the server requires additional explicit consent.

Text in your weclapp data, such as a customer's message in a ticket, can slip instructions to the AI (prompt injection). Our server marks free text from weclapp as untrusted, and changes run only through the preview. Grant write permissions only to people who can assess previews. Anthropic and OpenAI both point out that they do not vet servers you set up yourself; the trust decision is yours.

The AI works with the permissions of the weclapp account whose API key you stored, and can never do more than that account may do in weclapp. In weclapp, the associated user account appears as the author. You can store one shared key or one per person; then each person acts as themselves. With a shared account such as the mail inbox, that account appears as the sender.

Each person in wals.pro AI 4 weclapp has exactly one role (Viewer, Editor, Admin, area template or custom role). Roles can only restrict access, for example to "read only" or to certain data. Which license an API user needs, you clarify with weclapp beforehand.

Your data stays in weclapp. weclapp is certified to ISO/IEC 27001:2022, stores the data in the AWS Frankfurt region and, by its own statement, meets the requirements for GoBD-compliant software. Our server runs on Google Cloud in the EU (data center in Belgium), in data centers certified to ISO 27001:2022. You conclude the data processing agreement at registration; it names the sub-processors and governs data transfers to third countries. You can find it at ai.wals.pro/de/dpa. Responses from weclapp pass through and are not stored. What is stored: log metadata, approvals, a change record with target and document numbers (so that nothing is executed twice), usage counters and your credentials in the secret store, but not the contents of your records.

Requests to the AI provider run through its service and contract, possibly outside the EU; wals.pro operates no language model of its own. Because weclapp contains personal data, clarify with the data protection officer at your company whether your contract with the AI provider (plan, data processing) covers the processing.

Getting started is free permanently, within a monthly allowance; there are no credits. All features are free until December 31, 2026; after that there are paid tiers by scope. You pay for the AI client with the provider. Pricing models are explained in Per document or per user.