Claude connectors for Blender, Adobe and Fusion

What a connector actually is

A connector is a bridge between Claude and a piece of software. Under the hood it is an MCP server, which is a small program that publishes a list of things Claude is allowed to do inside that app, then carries out the ones Claude asks for. You are not copying answers out of a chat and doing them yourself. Claude reads the state of your open scene or file and makes the change in place.

That distinction matters for how you work. With a chatbot you ask "how do I batch rename 200 objects", get a Python snippet, and go paste it somewhere. With a connector you say "rename every object in this collection to match its material and tell me which ones had no material", and it happens in your file while you watch.

What shipped, in plain terms

Blender: set it up in 3 moves

1. 🖥️ Use the desktop app. Download Claude Desktop for macOS or Windows and sign in. The Blender connector talks to Blender over a local connection, so the browser version cannot run it. You also need Blender 4.2 or later from blender.org/download.
2. ➕ Add the connector in Claude. Open Claude Desktop, click Customize in the left sidebar, open the Connectors tab, find Blender, and select Add. This is the Claude half of the bridge and it takes about ten seconds.
3. 🧩 Install the Blender add-on. Open blender.org/lab/mcp-server in a browser next to an open Blender window and drag the "Drag and Drop into Blender" button onto Blender. You have to do it twice: the first drag adds the Blender Lab extension repository, the second installs the add-on. There is a download-and-install-from-disk option on the same page if drag and drop misbehaves.
4. ▶️ Start the server each session. Inside Blender go to Edit, then Preferences, then Add-ons, find the BlenderMCP panel and enable "start MCP server". Do this after you open your project. The add-on preferences also let you set the host, the port, and an auto-start option if you would rather not think about it again.
5. 💾 Remember that saving is still yours. Changes Claude makes apply to your live session and only hit disk when you save in Blender. Keep an incremental save habit while you are learning what it does.

Two prompts to run first. The first one is the honest test of whether a connector earns its place: hand it a mess you did not make.

You have access to my open Blender scene through the Blender connector. Before you change anything, walk the scene and give me a written map of it: the collections, the object count in each, the naming conventions people have actually used, the materials and which objects share them, and the modifier stacks that look heavy or duplicated.

Then flag the three things most likely to cause me problems later, for example unused data blocks, objects with no material, non-uniform scale that was never applied, or a node group that is doing something unnecessarily expensive. For each one, tell me what it is, why it matters when I render or export, and what the fix would be.

Do not modify anything yet. End with a numbered list of the fixes you would make, ordered by how much time each saves me, so I can pick which ones to approve.

Using the Blender connector, apply this cleanup pass across the objects in the collection I name, and work in one pass rather than object by object.

Rename every mesh object to the pattern type_name_variant in lowercase with underscores, using the material name where the object name is generic like Cube.001. Apply any non-uniform scale, clear parenting that points at deleted objects, and remove unused material slots. Leave anything ambiguous alone rather than guessing.

When you are done, report back as a table: the old name, the new name, and every change you made to that object. If you skipped an object, say which one and why. Then tell me the single Python operator I could add to Blender's interface as a button so I can run this same pass myself next time without asking you.

Adobe: describe the outcome, not the toolbar

1. 🔌 Connect it. In Claude (web or desktop) open Customize, go to Connectors, press +, browse the directory and connect "Adobe for creativity". This is the connector that fronts the 50+ Creative Cloud tools.
2. 🔑 Sign in with your Adobe account. You can start without one, but signing in unlocks higher usage limits, more tools, and work that saves across sessions, which is what you want the moment you are doing anything real.
3. 🗣️ Ask for the outcome. The whole point is that you do not choose the tool. You say what you need and Claude decides which Adobe tools to call, in what order, and with what settings. Naming an app in your prompt is fine as a hint, not as an instruction.
4. ✏️ Keep editing where you normally do. Output is not a dead end. You can push assets to Firefly Boards, open them in Express to keep iterating, or download them for the rest of your Adobe workflow.

Using the Adobe connector, take the image I am attaching and build me a full social pack from it, keeping one visual identity across every piece.

I need: a 1080x1350 Instagram post, a 1080x1920 story with the safe areas respected so nothing important sits under the interface, a 1200x630 link preview, and a 1500x500 banner. Recompose each one properly rather than cropping the same file four times, so the subject stays where the eye should land and the negative space still works at that ratio.

Match this direction: [describe your palette, your fonts, and the mood in one or two sentences]. Give me each file separately, tell me what you changed for each ratio and why, and flag anything that will look weak at small sizes so I can fix it before I post.

Using the Adobe connector, take the video I am attaching and repurpose it for vertical without me re-editing the whole thing.

Produce a 9:16 version that keeps the speaker framed properly through the cuts rather than a static centre crop, and a 1:1 version for feed. Keep the audio untouched. If reframing breaks a shot where something important sits at the edge, tell me the timestamp instead of quietly cropping it out.

Then give me a short list of the moments most worth clipping as standalone posts, with timestamps and a one-line reason each, based on where the point actually lands rather than where the energy is loudest.

Autodesk Fusion: text in, real geometry out

This is the one with the most plumbing, because Fusion runs its own local server that Claude then talks to. It is a five-minute setup and you only do it once.

1. ✅ Check the gate. You need an active Fusion subscription. Anthropic's own wording is that the connector lets designers and engineers "with a Fusion subscription" create and modify 3D models through conversation, so there is no free path here.
2. ⚙️ Turn on the Fusion MCP server. In Fusion, open Preferences, then General, then API. In the Preferences for Scripting and Programming panel, tick the Fusion MCP Server checkbox and note the Fusion MCP Server Port. The default is 27182.
3. ➕ Install the connector in Claude Desktop. Click Customize in the left sidebar, open the Connectors tab, press + to open the connector directory, search for Fusion, then choose Install next to Autodesk Fusion.
4. 🔁 Match the port. Toggle the connector to Enabled, click Configure, and confirm the port matches the one from Fusion's API preferences. A mismatched port is the single most common reason the connection silently does nothing.
5. 📄 Work in an open document. Claude acts on your active file, reading geometry and writing parametric history, so open the design you want to work on before you start prompting.

Using the Fusion connector on my open document, model this part with proper parametric history so I can edit it afterwards, and use user parameters for every dimension I am likely to change.

The part is: [describe it in plain language, with overall size, the features, and the tolerances that matter]. Create named user parameters for the driving dimensions rather than typing numbers into sketches, so I can change one value and have the model update.

As you go, tell me each feature you are adding and why in one line. When you finish, list every parameter you created with its value and what it controls, and flag any dimension you had to assume because I did not specify it.

Using the Fusion connector, review the model in my active document before changing anything, and tell me how it was actually built.

Walk the timeline and explain the design intent feature by feature in plain language: what each sketch and operation is doing, which dimensions are driving the shape, and where the model is fragile, meaning a change that would break downstream features or leave the geometry in a broken state.

Then give me a prioritised list of what you would fix to make it robust, for example under-constrained sketches, references hanging off faces instead of planes, or hard-coded numbers that should be parameters. Do not make the changes yet. I will pick which ones to approve.

Access is not skill

Every one of these connectors hands you capability. None of them hand you taste. iPhone cameras did not make everyone a photographer, they made the gap between people who understood light and people who did not much more obvious, because the excuse disappeared. Same thing here. When the software stops being the bottleneck, direction becomes the whole job.

Directing well is learnable and it is mostly one habit: say what the outcome is and what "good" looks like, then let the tool choose the route. Weak brief, "make this look better". Strong brief, "the eye should land on the label first, the background should read as evening not afternoon, and the whole thing has to survive being scaled down to a thumbnail". That second version is the part no connector writes for you.

1. 🎯 Pick the one connector closest to your actual work. Not the most impressive one. The one attached to the software you already open on a Tuesday, because that is the only one you will use twice.
2. 🧪 Run one real project through it, start to finish. A real deliverable with a real deadline, not a toy scene. Toy projects teach you the demo. Real ones teach you where it breaks.
3. 📝 Write down every correction you had to make. Every time you thought "no, not like that", that is a house rule you have been carrying in your head. Keep the list open while you work.
4. ♻️ Turn that list into a skill. Once you have five or six corrections, ask Claude to write them into a skill so you never repeat them. Install it under Customize, then Skills, then Add, then Upload a skill. Skills are available on every plan from Free through Enterprise.

I want to package my design standards into a reusable Claude skill so I stop repeating the same corrections every time I use the connectors.

Here are the corrections I found myself making this week: [paste your list]. Turn them into a SKILL.md with a short lowercase-hyphenated name, a one-line description saying exactly what it does and when it should trigger, and a set of clear rules written the way I would say them out loud.

Before you write it, ask me the three or four questions you need answered to make the rules specific rather than generic, for example my naming conventions, my export settings, and what I always want checked before I call something finished. Then write the skill and tell me how to save it as a zipped folder I can upload.

Proof

Claude for Creative Work, the official Anthropic announcement
Using the Blender connector in Claude, official tutorial
Blender MCP Server, install page and security warning
Adobe for creativity, Adobe's own announcement
Adobe for creativity, developer overview
Bringing Fusion onto Claude for Creative Work, Autodesk Platform Services
Use connectors to extend Claude's capabilities
When to use desktop and web connectors