AI is becoming a bigger part of engineering software, but there is an important difference between an AI assistant that explains how to do something and an AI agent that can actually perform the task inside engineering software.

Autodesk is taking this approach with Fusion Compute MCP, which is now available in public beta. With Fusion Compute MCP, AI agents can interact with Autodesk Fusion through the cloud and perform tasks such as creating parametric models, working with assemblies, preparing manufacturing operations, exporting files, and working with Fusion Team data.

What caught my attention is that this is not simply about asking AI to generate a CAD image or describe a modeling workflow. The agent can actually work with Fusion and create editable design data.

So, let’s take a closer look at how it works and what it can do.

What Is Fusion Compute MCP?

Before looking at Fusion Compute MCP itself, it is useful to understand MCP, or Model Context Protocol.

MCP is an open standard that allows AI applications to connect to external tools and data. Instead of an AI model only generating text, an agent can discover what tools are available and use them to perform actions. In the case of Fusion Compute MCP, Autodesk provides a cloud-based environment where an AI agent can interact with Fusion. According to Autodesk, the service exposes more than 7,000 endpoints through Fusion’s TypeScript API.

This means the agent can work with many of the operations that would normally be performed manually inside Fusion.

For example, an agent can:

• Create and modify sketches
• Create features such as extrusions, revolves, fillets, and chamfers
• Work with assemblies and joints
• Create manufacturing setups
• Generate CAM toolpaths
• Work with Fusion Team documents
• Export STL and STEP files

This makes the workflow very different from simply using AI as a chatbot.

From an Image to a Parametric CAD Model

One of the demonstrations from Autodesk shows an interesting example.

The agent is given a prompt to create a configurable, parametric wooden cabinet based on a reference image. Instead of producing only a visual representation, the agent creates the actual Fusion model. It creates the cabinet structure, shelves, doors, and drawers. The design can then be modified, such as replacing solid doors with glass panels and applying different appearances. The important part for me is that the result is not just a mesh or a rendered image. It is a native Fusion design with feature history and adjustable parameters.

That means the model can continue to be edited and configured after the AI has finished creating it. This is where AI-assisted CAD becomes more interesting for engineering workflows.

Designing a Part, Mold, and Toolpaths

Another Autodesk demonstration takes the workflow a step further.

The AI agent is asked to design a two-part injection-molded enclosure for a single-board computer.

The agent works through several design steps, including creating:

• A finned heat-sink lid
• A GPIO slot
• Screw counterbores
• Port cutouts
• The mold split

After creating the enclosure, the workflow continues into mold design. The agent creates a core and cavity, and then moves into CAM programming.

It creates operations such as:

• Face milling
• Adaptive roughing
• Parallel finishing

According to Autodesk, the toolpaths in this demonstration were generated in approximately 6.5 seconds. What I find interesting here is not simply the speed.

The important part is that the workflow connects design and manufacturing.

Instead of stopping after creating the CAD model, the agent can continue working toward manufacturing preparation.

Multiple AI Agents Working in Parallel

Fusion Compute MCP can also be used with multiple AI agents.

Autodesk demonstrates this using an autonomous robotic lawnmower containing approximately 1,100 colored parts. The task is to create five different color-layout options. Instead of creating each version sequentially, the workflow distributes the work across multiple agents. Four sub-agents each receive their own Fusion cloud instance and work on different versions of the design, while another version is handled by the main agent.

The resulting options can then be compiled into a comparison report. This approach introduces another interesting idea for engineering workflows:

AI agents don’t necessarily have to work one after another.

Different agents could potentially work on different design alternatives, configurations, or manufacturing tasks at the same time.

Three Fusion MCP Servers

Fusion now has three MCP-related services, and each one has a different purpose.

Autodesk Fusion MCP

The Fusion MCP works with the Fusion desktop application running locally on the user’s computer. It is focused on interacting with the local Fusion environment and performing design and make operations.

Autodesk Fusion Data MCP

The Fusion Data MCP focuses on the data surrounding the designs.

This includes areas such as:

• Hubs
• Projects
• Folders
• Permissions
• Properties
• Metadata

Autodesk Fusion Compute MCP

The new Fusion Compute MCP works differently. Instead of using the Fusion application running on the user’s computer, it runs Fusion sessions in the cloud. This means an AI agent can have its own Fusion environment to perform work without taking over the user’s desktop session.

For example, I could continue working on my computer while an AI agent works on another design task in its cloud Fusion environment.

I could then review the resulting model afterward.

What Can Fusion Compute MCP Do?

There are several areas where Fusion Compute MCP can interact with Fusion.

Parametric Modeling

The agent can create, read, update, and delete sketches and features.

This includes common operations such as:

• Extrude
• Revolve
• Fillet
• Chamfer
• Pattern

This is important because the agent is not limited to generating geometry once.

It can work with the parametric structure of the model.

Assemblies

Fusion Compute MCP can also work with assemblies.

Agents can create and manage components, occurrences, transformations, and joints.

This means AI interaction can extend beyond individual parts and into larger mechanical designs.

Manufacturing

The manufacturing side is another important part of the workflow.

Agents can work with:

• Manufacturing setups
• CAM templates
• Toolpaths
• Posted NC code

This creates a connection between CAD design and manufacturing preparation.

Documents and Export

The agent can also work with Fusion documents and Fusion Team projects. It can open and save documents, browse hubs and projects, and export files such as STL and STEP.

This makes the workflow more practical because the resulting engineering data can be stored and shared rather than remaining only inside the AI conversation.

The AI Model Still Matters

There is one important point to understand.

Fusion Compute MCP is the connection between the AI agent and Fusion. It is not the AI model itself.

The actual result depends on the AI model controlling the agent.

For example, the MCP can provide an agent with access to CAD and CAM operations, but the agent still needs to understand the design task, plan the workflow, select the correct operations, and interpret the results.

As AI models improve, the same Fusion Compute MCP environment may therefore become capable of handling more complex engineering tasks. This also means that AI-generated CAD should still be reviewed by an engineer.

A model can be created successfully from a prompt while still requiring changes to dimensions, design intent, manufacturability, tolerances, or other engineering requirements.

Public Beta Limitations

Fusion Compute MCP is currently available as a public beta, so there are some limitations.

At the moment, Autodesk provides:

• Two concurrent sessions per user
• A maximum session length of one hour
• Automatic session termination after 15 minutes of inactivity
• Queuing when the service reaches capacity

The sessions are also temporary.

If you create something that you want to keep, you need to save it to Fusion Team or export it before the session ends.

These limits are expected to change as Autodesk continues developing and scaling the service.

How to Get Started

To try Fusion Compute MCP, Autodesk says you need:

• An Autodesk account
• An active Fusion license
• A Fusion Team hub
• An MCP-compatible AI client

You then connect your MCP client to the Fusion Compute MCP service and ask the AI agent to start working with Fusion. The interesting part is that you don’t necessarily need to sit in front of Fusion and perform every operation manually.

The agent can work in its cloud Fusion environment while you review the results afterward.

What Does This Mean for CAD Workflows?

For me, the most interesting part of Fusion Compute MCP is the connection between AI and actual engineering operations. We have already seen AI become useful for generating ideas, explaining commands, writing code, and helping engineers find information.

Now the workflow is moving toward something different:

Ask → Design → Modify → Manufacture → Export

The AI agent can potentially move through multiple stages of the engineering process rather than stopping after generating a suggestion. Of course, this does not mean that engineers can simply give an AI a prompt and accept everything it produces.

The resulting geometry still needs to be checked. Dimensions, interfaces, design intent, manufacturability, tolerances, toolpaths, and final manufacturing requirements all still require engineering review. But the role of the engineer could change.

Instead of spending time performing every repetitive CAD operation manually, an engineer could increasingly focus on defining the design requirements, reviewing the results, and making engineering decisions.

Conclusion

Fusion Compute MCP gives us an early look at how AI agents could interact directly with CAD and manufacturing software.

Instead of AI simply telling us how to create a model, the agent can actually work inside Fusion, create parametric geometry, modify designs, build assemblies, prepare CAM operations, and export engineering data. The public beta is still limited, and the capabilities of the AI model driving the workflow remain an important factor. But the direction is clear: AI is moving from assisting engineers with information toward interacting directly with the tools engineers use to design and manufacture products.

It will be interesting to see how these workflows evolve as Autodesk continues developing Fusion Compute MCP and AI models become more capable.

• Author
• Recent Posts

ChampionXperience Team

Content Manager at ChampionXperience

Sharing insights, tutorials, and industry updates on CAD, simulation, engineering, and digital transformation technologies.

Latest posts by ChampionXperience Team (see all)

• Autodesk Fusion Compute MCP: Bringing AI Agents into CAD and Manufacturing - 21 September 2026
• Simcenter Flux 2026.1: 15x Faster EV Parasitic Extraction - 17 September 2026
• What’s New for Designers and Engineers with SOLIDWORKS Design AI - 16 September 2026