Turn your archive into an AI knowledge source with the Xtendis MCP connection

In a recent article, Wim Vis described how the new generation of Xtendis is building trust in an AI-driven world – with AI Search, anonymization, webhooks, and governance by design. Now, I zoom in on the next step: the Xtendis MCP Connection, which enables AI systems such as Microsoft Copilot, Google Gemini, and Anthropic Claude to communicate directly and securely with Xtendis.

When does AI become really valuable?

LLMs are impressive. They understand language, can reason, summarize, and much more. Yet, they lack exactly what an employee needs: knowledge about their own organization. What invoices still need to be approved? What is the status of a specific file? What was the correspondence with a particular client about?

AI only becomes truly valuable when those general capabilities – language, reasoning, summarizing – are combined with organization-specific knowledge. In a way that is secure, structured, and auditable. That's exactly what the Xtendis MCP Connection facilitates.

What is MCP?

MCP stands for Model Context Protocol: an open standard that allows AI systems to communicate in a structured way with external applications. More specifically, it means that an AI agent ‘understands’ what Xtendis is, what archives are available, how they are structured, and what document types, metadata, and workflows they contain.

The big advantage of MCP as an open standard is that you are not locked into one vendor or platform. Microsoft, Google, Anthropic, and Mistral all support the protocol. That fits seamlessly with how Xtendis has always looked at integration – and with the extensive API capabilities and previously introduced webhooks, too.

From search to action: four scenarios

During a recent webinar, we demonstrated exactly what is possible when an AI agent has access to Xtendis via MCP. The scenarios range from simple search queries to performing actions in the archive.

To see what the Xtendis MCP Connection looks like in practice, watch the full recording where we demonstrate the scenarios below (note: video is Dutch-only).

Search and retrieval with natural language

An employee types into Copilot, “Do I need to approve any invoices?”. The agent searches the central repository by document type, flow status, and authorizer – and provides a concrete answer. No separate search screen, you don't have to activate filters – simply ask the question. Exactly the type of interaction that we introduced before with AI Search is now also available via external AI assistants.

Analyzing and summarizing content

“What correspondence do we have with client 45636, and what was it about?”. The agent finds the documents and analyzes the text content to provide a summary. In the background, we are working on better full-text processing so that even handwritten and poorly scanned texts are reliably recognized – a requirement for AI to properly reason about document content.

Performing workflow actions

This is where things get really interesting. The agent is able to not just retrieve information, but also perform actions: approve an invoice, process a file, and add a note to a document. Because the connection runs via OAuth authentication using the agent's own user credentials, every action is logged in the audit trail. Xtendis' authorization and permission models remain in full force – even when the interaction runs through an AI agent.

Proactive thinking

The language model's memory functionality allows the agent to take context from previous sessions. Six cases are not complete? The agent suggests putting the next case on your name – just like during yesterday's session. In combination with the previously introduced webhooks, Xtendis can also proactively give an agent a command based on an event in the system.

Simplicity as a design principle

Everywhere we see organizations setting up complex AI projects. With good reason, AI is extremely suitable for complex problems. The real gain, however, is more in keeping those AI processes simple for the user. Not every employee needs to understand how an archive structure works, which search folder is the right one, or which metadata is available. They just need to be able to ask their question.

The MCP Connection unlocks the archive's complexity through a conversation. The next generation of employees – obviously raised on chat interfaces and AI – will have this as their basic expectation. Xtendis will soon be ready for that.

Governance remains the foundation

As with AI Search and anonymization, governance by design applies. AI may be the interface, but Xtendis remains responsible for authorization, security zones, retention policies, and anonymization of sensitive data. Every action performed through MCP goes through the same controls that an employee working directly in Xtendis encounters. That's not a concession – that is the way it should be.

Built together with customers

What we are showing in this ‘new generation Xtendis’ webinar series was not developed in a vacuum. It is largely based on interviews with customers: their challenges, their work processes, their expectations. Together, we are building the new generation of Xtendis: a sustainable platform that grows with new standards, growing information flows, and the opportunities AI offers. Updates and upgrades are included in the subscription – so you'll always be up to date.

What else is coming?

The MCP Connection is a first step that will be built upon. In a subsequent webinar, we'll explore more advanced scenarios: building dashboards via conversation, monitoring requests, and linking to the other AI initiatives we're working on.