MCP Is Turning LoyJoy Agents Into Multi-Talents

Since OpenAI released remote MCP at the end of May 2025, we have been working enthusiastically to integrate LoyJoy into the MCP ecosystem. LoyJoy masters the Model Context Protocol, both as a client and as a server.

Large language models now access LoyJoy processes, modules and analytics directly via lightweight streamable HTTP endpoints. And vice versa, external MCP services can be integrated into any chat via drag-and-drop.

What is MCP actually?

MCP (Model Context Protocol) is an open standard introduced by the Claude makers at Anthropic at the end of 2024.

It works like a USB port between AI models and external data sources: A language model can use simple endpoints to retrieve context-relevant information or trigger functions - without proprietary integration logic.

Why is MCP a game changer?

• Federated agents - any software can be remotely controlled by AI, becoming an AI capability.
• Faster time-to-value - no custom code, no long sprints for integrations. Instead, the AI handles integrations on an ongoing basis.
• Governance & GDPR - data remains in LoyJoy’s EU cloud.

First role: LoyJoy as an MCP client

URLs from external MCP servers (e.g. SAP functions or proprietary knowledge sources) can already be integrated into LoyJoy agents.

Each linked URL becomes a capability for the agent: ‘Calculate the price → write an invoice → store it in SharePoint’.

Second role: The LoyJoy chat agent as an MCP server

All process modules of a BPMN flow in LoyJoy will automatically become ‘MCP tools’ in future. At the same time, we compress the BPMN model and give it to the AI as a ‘happy flow’.

And if the happy flow is not enough? The exceptions to the happy flow are managed by the AI! The AI then independently triggers LoyJoy module calls in the right places. Without any complex, susceptible prompt engineering.

Third role: The LoyJoy Manager as an MCP server

You already have read access to process models and analytics data via MCP. In the near future, you will be able to remotely control the LoyJoy Manager (our backend) from your personal AI.

For example, your ChatGPT will be able to answer questions about LoyJoy in the browser. Example: ‘How did the conversion rate of my LoyJoy chatbot develop last week?’. The answer is based directly on the data from the LoyJoy KPI dashboard - with role-based security, of course.

Or you can simply describe which AI agent you would like to build in LoyJoy. The AI then assembles it independently with the appropriate modules.

Outlook: Monoliths are dissolving, phonebots and avatars are coming

MCP is a new architectural pattern for connecting AI and existing systems. It dissolves the RAG pipelines that have often grown monolithically in recent years, in which the knowledge database, model and UI were closely coupled.

Instead - almost like USB - systems, data sources and large language models can be plugged together as required.

The LoyJoy chat agent as an MCP server opens up completely new usage patterns - from voicebots on the phone to photorealistic video avatars that all address the same MCP server.

— by Laura Knittel