AI tools for Mendix, built for coding agents
AIDE Pro is an agentic development tool for Mendix Studio Pro. Connect Claude Code, Cursor, GitHub Copilot, Antigravity or any MCP client and it gets 243 tools to build real microflows, entities, pages and security - then verify its own work against your app model.
- 243
- Model tools
- MCP
- Any agent
- 0
- Keys stored
- Claude Code
- Cursor
- GitHub Copilot
- Any MCP client
Still on Mendix 10 LTS? That edition is free to download today.
What are AI tools for Mendix?
AI tools for Mendix let an AI model read and change a Mendix application’s model - its entities, microflows, pages and security - instead of only writing text about it. That distinction matters on Mendix more than on most platforms, because a Mendix app is not a folder of source files but a structured model. An agent that can only emit code has nothing useful to emit, and an agent that writes to the model file directly bypasses every check the platform would have applied.
AIDE Pro closes that gap. It is an extension for Mendix Studio Pro 11 that exposes 243 model tools over the Model Context Protocol (MCP), so the AI coding agent you already use can create and edit real Mendix artifacts, run the app, and check its own work. AIDE Pro calls no AI model of its own: your agent brings the model and the credentials, and AIDE Pro executes the model operations.
In short: your agent gets hands, not just opinions. It writes to the actual Mendix model, and the platform’s own consistency engine confirms the result before you ever open the app.
Claude Code, Cursor, Copilot, Antigravity, or your own client.
AIDE Pro is an MCP server for Mendix. If your agent speaks MCP, it can drive Studio Pro - there is no new chat window to learn and no second subscription to buy.
Claude Code for Mendix
Point Claude Code at the AIDE Pro bridge and it can read your whole app model and build in it. Claude Code drives long, multi-step Mendix work well because it plans, calls tools in sequence, and reacts to the consistency errors it gets back.
Cursor for Mendix
Add AIDE Pro as an MCP server in Cursor and you can code in Mendix from the editor you already use. Cursor keeps your prompts and rules; AIDE Pro supplies the Mendix model operations it otherwise has no way to perform.
GitHub Copilot, Antigravity and others
Nothing about AIDE Pro is tied to one vendor. It advertises its tools over standard MCP, so GitHub Copilot, Antigravity, a custom in-house agent, or whatever ships next all connect the same way.
Vibe coding in Mendix: why a chat window isn’t enough
Vibe coding works on text codebases because the model writes the artifact directly. On Mendix it usually stalls, and the reason is structural rather than a failure of the model.
See how the tools workA Mendix app is a model, not files
Ask a chat assistant for a microflow and you get a description of one. Someone still has to click it into Studio Pro, which is where the time actually goes.
Editing the file directly is how projects break
A generic agent CAN open your .mpr, and that is the risk rather than the reassurance. Writing to it directly bypasses every validation the platform applies, and one plausible-looking write can leave the project unable to open at all.
Plausible is not the same as valid
Mendix has strict consistency rules. Generated logic that reads well can still be unbuildable, and you only find out later, by hand.
What AIDE Pro changes
The agent calls real model operations, so the artifact exists when the call returns. Then it runs the consistency check and reads the result back, so a broken change surfaces in seconds rather than at your next build.
243 tools your agent can call.
A development surface, not a chatbot bolted on the side. Every group below is a set of real model operations.
Build the domain model
Entities, attributes, associations including cross-module, enumerations, indexes and validation rules - created and read back.
Author microflows
Activities, decisions, loops, merges, error handlers and REST calls - with the platform’s own consistency engine confirming each step.
Compose pages
Data views, list views, widgets, tab containers, layouts and captions - the real page tree, verified after every write.
Model security
User roles, module roles, page and microflow access, and additive entity access rules with row-level XPath isolation.
Run & test live
Build and start the app, then query it with live OQL and read the runtime log - close the loop without leaving the model.
Analyze & audit
Read-only analyzers for security gaps, dead code, performance, concurrency and data-model quality across the whole app graph.
Integrate
JSON structures, REST calls, import and export mappings, and OQL-backed view entities - wired end to end, not stubbed.
Keep the project tidy
Folders, document moves, layout tidying and renames - so a generated module still looks like something a person would hand over.
How the agent proves its work.
This is the part that separates a useful agent from a fast one. AIDE Pro makes the agent check itself, in the platform, before it reports success.
It writes to the model
The tool call creates the entity, microflow or page for real. When the call returns, the artifact exists in Studio Pro.
It reads the change back
Every write is followed by a read of what was actually stored - not a restatement of what was requested. Silent property drops surface here.
It runs the consistency check
Mendix’s own error list, the same one Studio Pro shows you. Unbuildable logic is caught immediately, with the error codes, so the agent can fix it in the same session.
It queries the running app
With the app running, the agent can execute live OQL and read the runtime log - so “the data looks right” becomes a query result instead of an assumption.
Where we are deliberately blunt: a consistency-clean model is not the same as a correct application. AIDE Pro shortens the feedback loop and removes whole classes of error; it does not remove the need for a developer to review the work. Our documentation says so too, including a list of known limitations.
AIDE Pro vs an AI chat assistant vs a generic coding agent.
All three can help. Only one of them changes your Mendix model and then checks the result.
| Capability | AIDE Pro | AI chat assistant | Generic coding agent |
|---|---|---|---|
| Reads your whole app model | Yes - every module, entity, microflow and page | Only what you paste in | Only as raw database rows, with no model meaning attached |
| Creates real Mendix artifacts | Yes - 243 model tools, through the platform’s own APIs | No - it describes them | Only by writing to the model file directly, unvalidated |
| Verifies its own work | Read-back, consistency check, live OQL | No | No |
| Risk of corrupting the project | Writes go through Mendix’s own validated, transactional APIs | None, because it changes nothing | Real - a bad direct write can stop the project opening |
| Runs and queries the app | Yes | No | No |
| Stores API keys | None - it calls no model | Typically yours | Yours |
| Works with your existing agent | Any MCP client | Its own window | Not applicable |
| Learns your project standards | Yes - skills and learnings travel with the app | Per conversation | Per repository |
Nothing to leak, because nothing is stored.
AIDE Pro does not call an LLM and holds no credentials. Your agent owns the model and the keys; the extension only executes model operations on the machine it runs on.
No API keys, ever
AIDE Pro stores no provider keys and has no model settings. Your agent brings its own credentials.
Loopback-only bridge
The tool bridge binds to localhost and refuses browser requests. Nothing is exposed off the machine.
You choose the model
Run a local model through your agent for a fully private setup, or a managed cloud model. That is your decision, not ours.
No telemetry
AIDE Pro never phones home. Your model, prompts and code are not collected.
Never trains on your code
Your application and data are never used to train any model.
Runs where you run
On your workstation or a controlled build environment. The model layer is wherever your agent points it.
What it changes economically.
We would rather you worked this out with your own numbers than trusted a percentage we made up.
- You are not buying another AI subscription. The agent and the model are yours, so there is no per-seat AI cost layered on top of this and no lock-in to a model we picked. If your team already pays for Claude Code or Cursor, that spend starts doing more work.
- The saving is rework, not typing. The expensive hours in Mendix delivery are senior hours spent clicking out decisions already made, and redoing work the consistency check rejects later. AIDE Pro moves that rejection to seconds after the write, while the context is still in someone’s head.
- Knowledge compounds instead of resetting. Skills and learnings record how your project is built and travel with it, so the second application does not re-pay the first one’s learning curve, and a new developer inherits the standards rather than guessing them.
- The licence may not be the cost you expect. There is a route to a no-cost Mendix 11 licence for teams we work with, so the sensible first step is to ask rather than to budget for a price you have guessed at.
Work it out on your own project. Take your senior Mendix day rate. Count the hours a sprint goes into building CRUD pages, wiring widgets to data, and redoing work the consistency check throws back. That is the surface this compresses. We will do that calculation with you on a call, using your figures, and tell you if it does not add up for your situation.
How to set up AI-assisted Mendix development.
Three steps, and you keep the agent and editor you already use.
Install AIDE Pro into your app
AIDE Pro installs per Mendix app, alongside your project. We provide the distribution and the setup guide for your environment.
Connect your agent over MCP
Add the local AIDE Pro bridge to Claude Code, Cursor, GitHub Copilot, Antigravity or your own client. It advertises all 243 tools on connection.
Give it work, and let it verify
Describe the change. The agent builds it in the model, checks consistency, and reports what it actually did - with the error list if there is one.
Two editions, depending on your Mendix version.
Everything described on this page is the Mendix 11 edition. If you are still on Mendix 10 LTS, the earlier edition remains available.
| Mendix 11 edition | Mendix 10 LTS edition | |
|---|---|---|
| What it is | The agent edition described on this page - 243 model tools exposed to your own AI coding agent over MCP. | The earlier release, for teams still on the Mendix 10 long-term-support line. |
| How to get it | Fill in the form and our team will get you set up. | Download from the Mendix Marketplace, then request a licence key. |
| Cost | There is a route to a no-cost licence for teams we work with. Ask how it applies to you. | Free, with a licence key. |
| Requires | Mendix Studio Pro 11 and an MCP-compatible agent. | Mendix Studio Pro 10 LTS. |
Ask about a free Mendix 11 copy
Still on Mendix 10 LTS? The earlier edition is free to download, and you can request a licence key for it.
Download for Mendix 10 (opens in a new tab)
Get a Mendix 10 licence key (opens in a new tab)
What you get besides a licence.
AIDE Pro is built by a team that delivers on Mendix every day, and it is shaped by that work: the tools exist because we needed them on real projects, and the limitations are published because we hit them first. That is the part worth checking before you commit to anything. We would rather tell you where it stops than have you discover it in a sprint.
So the engagement is not a handover. We walk it through on a real application, help your team fit it to how you already work, and stay reachable while they get used to it. If it turns out not to suit your situation, we will say so on the call. Mendix + AI · Mendix Studio Pro · AIDE Pro docs
AI development on Mendix, answered.
What are the best AI tools for Mendix?
Can I use Claude Code with Mendix?
Can I code in Mendix with Cursor?
Can you vibe code a Mendix app?
Is there an MCP server for Mendix?
How is AIDE Pro different from an AI chat plugin for Mendix?
Does AIDE Pro work alongside Mendix Maia?
Does AIDE Pro store my API keys or send my code anywhere?
Which Mendix version does AIDE Pro need?
Is AIDE Pro free?
How do we justify the cost internally?
How do I get a free copy of AIDE Pro for Mendix 11?
Put an AI agent to work on your Mendix app.
Tell us about your project, your Mendix version, and the agent your team already uses. You get a walkthrough on a real application, an honest account of what the tools do not do yet, and a team that stays reachable while yours adopts them.