Claude Now Connects to Autodesk Fusion
With the Autodesk Fusion connector, designers and engineers can turn natural language into design actions, iterate without starting over, and stay in flow throughout the design process.
This lesson is original educational writing based on this video by Anthropic (published April 28, 2026). All credit for the original content goes to the creators.
The Gap Between Design Intent and CAD Execution
Anyone who has worked with professional CAD software knows the experience: you have a clear mental model of what you want to create, and the execution bottleneck is not creativity or engineering judgment — it is navigating the software. Modern parametric CAD tools like Autodesk Fusion are extraordinarily capable, but that capability comes at the cost of complexity. Menus nest several levels deep. Features have dependencies that must be defined in the right order. Parameter names are specific and must be recalled precisely. Operations that seem conceptually simple — “make this wall thicker” — may require navigating through the feature tree, editing a sketch, updating a dimension, and resolving any downstream dependencies that the change affects.
This is not a complaint about bad software design. The complexity is to some extent inherent to the problem: parametric 3D modeling is a genuinely complex domain, and the software that handles it well is necessarily sophisticated. But the sophistication creates a real barrier for users who know mechanical engineering but are not CAD power users, for people who are trying to iterate quickly on design concepts, and for anyone who finds themselves spending more time fighting the interface than thinking about the design.
The Claude-Autodesk Fusion connector addresses this barrier by adding a natural language interface on top of Fusion’s existing capabilities. Rather than navigating menus to achieve an intent, you can describe the intent and Claude translates it into the appropriate Fusion operations.
What Claude Can Do in Fusion
The connector exposes a set of Fusion capabilities that Claude can invoke through natural language instructions. Understanding the scope of what is available helps you recognize when to use the connector versus when to work directly in Fusion.
Creating and modifying features is the most common use case. Claude can create standard Fusion features — extrusions, pockets, fillets, chamfers, holes, shells — from natural language descriptions. “Add a 5mm fillet to all the outer edges of this bracket” is a representable instruction. “Create an M8 threaded hole 20mm deep in the top face” maps to specific Fusion operations Claude can execute without requiring you to navigate the Hole feature dialog manually.
Modifying parameters is particularly valuable for parametric design workflows. Fusion’s strength is the ability to drive geometry from named parameters — if you set up your model correctly, changing a parameter value updates all dependent geometry automatically. Claude can identify and modify named parameters from natural language: “change the wall thickness to 3mm” can be executed as a parameter update if wall thickness is a named parameter in the model.
Running design checks is another valuable operation type. Claude can invoke Fusion’s analysis tools — checking wall thickness against minimum manufacturing thresholds, identifying faces that violate design rules, or verifying clearances between components — and report the results in natural language. Instead of running each analysis tool separately and interpreting the output, you can ask “Does this part have any features that would be problematic for injection molding?” and get a synthesized response.
Iterating on design concepts is where the connector most clearly changes the workflow character. Traditionally, design iteration in CAD requires making a change, evaluating it, and then either proceeding or undoing and trying something different. With the connector, you can describe alternatives conversationally: “try making the base plate 10% wider and see if that improves the stability calculation” — Claude makes the change, reports the result, and you can continue the conversation to decide whether to keep it or try a different approach.
How This Changes the Design Workflow
The traditional CAD workflow is software-mediated and menu-driven. You think in terms of operations — which feature to add, which parameter to change, which dialog to open. The cognitive model is structured around what the software can do and how to invoke it.
The connector shifts this toward an intent-mediated workflow. You think in terms of what you want the design to achieve, and the translation from intent to software operation is handled by Claude. For experienced CAD users, this does not eliminate the value of deep Fusion knowledge — understanding what operations are possible and how parametric dependencies work still matters. But it removes the friction of menu navigation for operations you know should be straightforward.
For less experienced CAD users — engineers who understand the design requirements but are not Fusion power users — the impact is more significant. They can now attempt operations that they know are theoretically possible in Fusion without needing to know exactly which menu path to follow. The cognitive barrier between “I know what I want to achieve” and “I can execute this in the software” drops considerably.
The MCP Protocol Underlying the Connector
The Autodesk Fusion connector is built on the same Model Context Protocol infrastructure as Claude’s other application integrations. MCP defines a standard way for Claude to discover what operations a connected application supports, how to call them, and what parameters they require.
From Autodesk’s perspective, building the Fusion connector meant implementing an MCP server that exposes a defined set of Fusion API operations through the protocol’s standard interface. Claude connects to this server when the integration is active, queries what operations are available, and uses that knowledge to map natural language instructions to specific API calls.
This architecture has an important implication for what Claude can and cannot do in Fusion: Claude is limited to the operations that the MCP connector exposes. Operations that Autodesk did not include in the connector cannot be invoked through Claude, even if they are fully supported in Fusion’s own API. The connector is a curated exposure of Fusion capabilities, not a complete API wrapper.
As Autodesk expands the connector over time, the range of operations available through natural language will grow. This is typical of MCP integrations in their early versions: they start with the most common and safely automatable operations and expand as the integration matures and edge cases are better understood.
What Still Requires Manual Work
Understanding the limits of what the connector handles helps you use it effectively rather than being frustrated when natural language instructions do not produce the expected result.
Complex surfacing and freeform modeling remain primarily manual activities. These involve visual and spatial judgment that is difficult to express in natural language at the precision Fusion requires. Describing a complex curved surface in words rarely produces the exact geometry you have in mind.
Assembly constraints and motion studies can be partially addressed through the connector but often require visual inspection and manual adjustment. Getting joints right, verifying clearances in motion, and tuning simulation parameters are areas where direct interaction with the Fusion viewport remains more efficient.
Post-processing and output configuration — setting up CAM toolpaths, configuring rendering materials, preparing drawings with specific annotation standards — involves detailed decisions that often benefit from direct interface interaction rather than natural language description.
Model debugging and repair — fixing import errors, resolving failed features in complex feature trees, recovering from failed operations — often requires navigating the feature tree and understanding the specific failure mode visually. Claude can help diagnose problems, but the repair may still require manual steps.
Check your understanding
3 questions · your answers are saved in this browser only
-
1. What is the primary problem the Fusion connector is designed to solve?
-
2. What limits which Fusion operations Claude can perform through the connector?
-
3. For which type of task does the Fusion connector provide the least value?
Build it yourself
Follow these exact steps to reproduce it yourself
Build It: Set Up and Test the Fusion Connector
Step 1: Enable the Autodesk Fusion connector in Claude
In Claude.ai settings, navigate to Integrations. Find the Autodesk Fusion connector and follow the authorization flow — you will need an active Autodesk account with Fusion access. Complete the OAuth authorization that allows Claude to interact with your Fusion workspace. Note what permissions the connector requests; this tells you what operations it can perform.
Step 2: Create a test part and practice basic natural language operations
Open a simple mechanical part in Fusion — either an existing part you know well or a simple bracket you create from scratch. With Claude open alongside Fusion, try the following natural language operations in sequence, noting which execute correctly and which require clarification:
- “Add a 2mm fillet to all outer edges”
- “Create a 6mm diameter through-hole in the center of the top face”
- “Check the minimum wall thickness of this part and tell me if any areas are below 1.5mm”
For each operation, compare the result to what you would have achieved navigating Fusion manually. Note the operations where the connector understood your intent precisely and any where it needed follow-up clarification.
Step 3: Use the connector for a real design iteration cycle
Take a parametric model where you are exploring design alternatives. Use Claude to run at least five iterations: describe each variant in natural language, have Claude implement it, evaluate the result, and ask for the next variation. This might look like:
“Make the base 10% wider and see how that changes the mass” “Revert that and instead add a second mounting hole on the left side” “Increase the wall thickness from 3mm to 4mm and re-run the wall thickness check”
Document which steps benefited most from natural language and which you needed to handle manually. This creates a personal map of where the connector adds most value in your specific workflow.