Mendix AI agent integration for your own service and order apps

What is a Mendix AI agent integration?

It is a connection through the services your Mendix app publishes, so an AI agent can read data and, where you allow it, write back. A customer reports a fault; the agent asks for the missing details and puts them in the right fields of the ticket in your Mendix app, for your service desk.

Mendix apps integrate with other systems through REST, OData, GraphQL, SOAP, Change Data Capture and business events. For an agent, published REST services and published OData services are the usual route: both expose entities and microflows to other applications.

What does this look like in practice?

At Unica, BeeManaged is piloting a service agent that collects missing details from customers and writes them straight into the ticket in Mendix, Unica's own system. The service desk sees the status change from 'AI enrichment active' to 'AI complete'. The process starts and ends in Mendix, and Unica stays in charge.

• Case: Unica service agent with Mendix

What does Mendix offer itself for AI?

Mendix has AI for building apps and modules for building agents inside your app. Maia is the AI assistance in Studio Pro for developers. The Agents Kit is a collection of starter apps, connectors and modules to create agents that work with your app's data and logic, with models from Mendix Cloud GenAI, OpenAI, Microsoft, Mistral, Gemini or Amazon Bedrock.

That route suits teams that build and maintain the agent themselves in Mendix. A BeeManaged agent sits outside the app: it combines your Mendix app with your ERP, CRM, website and WhatsApp, and we manage it.

• Maia: AI assistance for development in Studio Pro; Maia does not use project or customer data for model training, and each company gets 2,400 free Maia units per month.
• Agents Kit: modules such as Agent Commons, with human-in-the-loop controls, GenAI Commons and Conversational UI for a chat in your app.
• Mendix Cloud GenAI resource packs: a premium product with Anthropic Claude and Cohere models, available in the region Mendix Cloud EU (Frankfurt), with the models running through cross-region inference within European AWS regions.

Can a Mendix app be an MCP server?

Yes. Mendix provides an MCP Server module that makes microflows in your app available as tools for AI agents, over Streamable HTTP. Mendix says the module is still in an early version that may include breaking changes, and strongly recommends keeping the user in the loop, for example with a confirmation step, for tools that act in the real world, such as sending email.

• Authentication runs through a microflow that returns a Mendix user, so entity access applies. Without authentication, anyone can reach the server.
• Tools can only return text.
• The Studio Pro MCP server is something else: a local tool for developers to let coding agents work on the app model.

• What an MCP server is and what it does not arrange

Agents Kit, MCP server module or a managed agent: which fits?

Use the Agents Kit when your Mendix team wants to build an agent inside the app. The MCP Server module suits exposing app logic to AI assistants your organisation already uses. Choose a managed agent when customers talk to it on your website or WhatsApp, the answer needs other systems as well, and you want it built and managed for you.

Three routes compared (as of October 2026)

How does the connection with a Mendix app work technically?

Your Mendix developer publishes a REST or OData service with the entities and microflows the process needs. The agent calls it with credentials that map to a Mendix user, so module roles and entity access rules decide what it can read and change. With OData v4 you set per entity whether it is insertable, readable, updateable and deletable.

• Authentication of published services: basic authentication, active session or custom authentication through a microflow that returns a user. The OIDC SSO module adds OAuth access tokens, including the client credentials grant for machine identities.
• Entity access rules apply to the service's user. If the agent passes an object to a published microflow with an attribute that user may not write, the request fails.
• On Mendix Cloud you can restrict paths such as the service endpoint by IP range or client certificate.
• Hosting: Mendix Cloud runs on AWS, with European locations in Ireland and Frankfurt; Mendix on Kubernetes runs the app in your own private cluster.
• We use a dedicated user for the agent with only the module roles it needs, and record every question and action in the audit log.

How we set it up, in four steps

1. 01

   Choose one process

   For example completing incoming tickets, with the fields and values your Mendix app accepts.

2. 02

   Publish a minimal service

   Your Mendix developer exposes only the entities and microflows the process needs, with a dedicated user and module role for the agent.

3. 03

   Go live with approval

   Changes wait in the approval queue by default, or go straight into the app where you ask for that, as at Unica.

4. 04

   Monitor and improve

   Conversation review, audit log and a monthly report with the themes that came up, then the next process.

• How BeeManaged works, step by step

Pricing for a Mendix agent

Customer-facing agents cost from €0.35 per conversation. Setup is quoted up front (development €175 per hour), monitoring and the dashboard cost €250 per month, and a consultation package is €1,750. Internal agents and ticket processing are custom work, priced in consultation. Work in the Mendix app itself is done by your own Mendix developer or partner.

• All BeeManaged prices

When don't you need BeeManaged for Mendix?

If your Mendix team wants to build and run an agent inside the app for its own users, the Agents Kit gives them the building blocks. BeeManaged makes sense when customers or dealers talk to the agent outside the app, the answer needs other systems as well, and you would rather not build and manage the agent yourself.