The Trending Blender AI Add-on Has No Code in Its Repo. Check This Before You Install Any.
A one-minute check for any AI or MCP add-on you find on a trending list, using a real example from this week, plus the safer way to get Claude or ChatGPT driving Blender.
AI add-ons are the easiest software in the world to sell right now. Promise that a chatbot will model, light and texture your 3D scene, put "MCP" in the name, and people click download.
That is exactly why the add-on at number 18 on Trendshift's daily GitHub board this Saturday (November 21, 2026) deserves a closer look. Trendshift is a site that ranks the GitHub repositories gaining attention fastest. The repo is called bookwjitchieni1986/blender-ai-mcp-addon, it promises an AI assistant and an MCP server inside Blender, and it shows about 2.2k stars. Open it and you find a README, a license file and a docs folder. No add-on code at all.
Nobody here has opened the ZIP it offers, so this article does not claim it is malware. It claims something more useful: the warning signs were visible in under a minute, from the repo page alone, and the same minute will protect you from the next one.
What the add-on promises
First, two terms. Blender is the free, open-source 3D creation suite used for modeling, animation and rendering. MCP (Model Context Protocol) is the standard way AI apps like Claude Desktop or Claude Code connect to outside tools. An MCP server for Blender lets an AI model read your scene and run commands inside it.
That idea is real and popular. The project most people mean when they say "Blender MCP" is ahujasid/mcp-for-blender (formerly blender-mcp), an MIT-licensed community plugin with roughly 35k stars.
The trending repo's About text offers a bigger version of the same thing: "AI assistant and MCP server for Blender: visual chat, scene tools, multi-agent workflows. Works with ChatGPT/GPT, Claude, Gemini, Ollama, OpenCode." Its README lists scene tools, a studio window, and a tool called execute_code.
To use it, the README tells you to download a ZIP from the Releases page and install it in Blender. The release notes say the bundle targets Windows x64 and includes a "Portable Python 3.13 runtime." So you would be installing prebuilt files whose source you cannot read, into an app where add-ons run Python with your user's permissions.
What the repo page shows
Here is what turned up reading the repo on November 21, in the order a careful reader would notice it.
There is no code. The repo root holds three items: docs, LICENSE and README.md. Compare that with the original project, whose root has addon.py, a src/blender_mcp package, tests, a pyproject.toml and install scripts. A real add-on repo has the add-on in it.
The account is a week old. The owner, bookwjitchieni1986, joined GitHub on November 13, 2026 and has one repository. The only release, v1.0.0, went up on November 19. The repo showed about 2.2k stars, 32 forks and 1.1k watchers two days later. That is a very fast climb for a first project from a brand-new account.
The license names someone else. The LICENSE file is MIT, with the line "Copyright (c) 2025 Siddharth Ahuja." Siddharth Ahuja is the author of the original Blender MCP project. Reusing MIT code is allowed. Shipping a license with another person's name while offering no code to compare against it is a mismatch worth noticing.
The instructions don't agree with each other. The README says to download mcp-for-blender-pro.zip and install it through "Get Extensions." The release notes say to download blender-ai-mcp-addon-v1.0.0.zip and install it through "Add-ons." The README's heading calls the project "mcp-for-blender-pro," which borrows the original's new name, mcp-for-blender.
The README is written for search engines. It ends with a keyword block of about 40 terms ("blender mcp server," "blender chatgpt," "mcp-for-blender" and so on). Genuine projects rarely need one.
The README is honest about one thing. It advertises an execute_code tool and says its Python tool validation "helps catch disallowed operations; it is not a security sandbox." If the shipped add-on works as described, a model, or anyone who can steer the model, could run Python on your machine through Blender. Nobody can confirm that from the repo, because the code isn't there.
Any one of these signs has an innocent explanation. A developer might keep source private, reuse a license file carelessly or write messy docs. All of them together, on a repo that asks you to run a prebuilt Windows bundle, fits the shape of a repackaged lure: borrow a famous project's name, add fast stars, ship a ZIP. Treat it that way until someone shows otherwise.
The one-minute check
This works for any AI add-on, MCP server, agent skill or plugin you find on a trending list or in a search result. You need a browser, or the GitHub CLI (gh) if you prefer a terminal.
1. Is the code in the repo? Look at the root listing. You want source files you could read, not only a README and a release download.
gh api repos/OWNER/REPO/contents --jq '.[].name'
2. Who made it, and when? A first-ever repository from an account created days ago is not proof of anything, but it raises the bar for everything else.
gh api users/OWNER --jq '{created: .created_at, repos: .public_repos}'
gh repo view OWNER/REPO --json createdAt,stargazerCount
3. Does the license match the owner? Open the LICENSE file and read the copyright line. If it names a different person, find that person's project. It is often the one you actually wanted.
4. Do the install steps agree? Compare the README with the release notes. File names and menu paths that differ are a sign nobody tested the instructions, or that one of them was copied from somewhere else.
5. Is there an original? Search for the project's core name. If a much older, much larger project does the same thing, use that one unless you have a specific reason not to.
6. What does it run, and with what access? Look for tools that execute code, shell commands or network calls. If the project itself says it is "not a security sandbox," believe it.
If a repo fails two or more of these, don't install it on the machine that holds your API keys, client files or browser sessions.
The safer way to get AI driving Blender
If what you wanted was Claude or another model controlling Blender, install the original project from its own repo. Its README lists Blender 3.0 or newer, Python 3.10 or newer and uv (a fast Python package runner) as prerequisites, then lays out a manual path (uvx is uv's command for running a Python tool without installing it permanently):
# 1. Register the MCP server with Claude Code
claude mcp add blender uvx mcp-for-blender
# 2. Install the Blender add-on
uvx mcp-for-blender install-addon
Then in Blender, open Edit, Preferences, Add-ons, and enable "Interface: MCP for Blender." The add-on source (addon.py) sits in the repo where you can read it before you run it, and the Python package installs from PyPI under the name mcp-for-blender (the older uvx blender-mcp command still works, according to the README).
The original project also offers one-line installers that pipe a script from its website into your shell (curl ... | sh on macOS and Linux, irm ... | iex on Windows). They are convenient. They also run whatever that URL serves at that moment, so the manual path above is the one to use if you want to see each step. The project also sells a premium tier through its website, with its own terms file in the repo; the MIT license covers the open code, so read the terms before you sign up for anything extra.
Honest limitations
Start with what this article cannot tell you. Nobody here downloaded or opened the trending repo's ZIP, so there is no evidence it contains anything harmful. The signs above are signs, and the owner could turn out to be a developer who simply prefers to ship binaries. Release asset sizes and file lists were not readable from the release page.
The original project has its own risks. Its README states that the add-on's socket server "has no authentication or encryption," which means any program on your machine that can reach that local port can send Blender commands. Run it only while you are using it. Any MCP setup that lets a model execute Python inside Blender gives that model, and any text that steers it (a downloaded asset description, a prompt in a file), the ability to run code as you. That is the price of the feature, not a bug in one project.
Everything here is a snapshot. The trending repo and its release can change at any time (nothing ties it to a fixed commit, the frozen snapshot of a repo that can't be edited afterward), so a reader next week may see different files, counts or instructions. Run the check yourself rather than trusting this article's description.
Star counts are a weak signal in both directions. A brand-new repo can buy or farm them, and a solid small project may have very few. Treat stars as a reason to look, never as a reason to trust.
And this check takes a minute because it is shallow. It catches the lazy lures. A careful attacker who publishes real-looking source and a matching binary will pass it, which is why the habit of preferring the original, and installing from source you can read, matters more than any single checklist.
What to do next
Before your next AI add-on install, run the six questions. If you already installed something from a trending list that fails them, remove it, rotate any API keys you entered into it, and check what else that machine had access to.
The tools that let models act on your computer are going to keep multiplying, and every popular one will get copycats. The minute you spend reading the repo is the cheapest security you will ever buy.
Sources: bookwjitchieni1986/blender-ai-mcp-addon repo, README, LICENSE, releases (v1.0.0, Nov 19 2026) and owner profile, all read Nov 21 2026; ahujasid/mcp-for-blender README; Trendshift daily board, read about 15:15 ET Nov 21 2026; GitHub CLI gh api manual.