Hermes Agent Blender gets an 7.9 out of 10 from me: the official plugin is the cleanest way yet to let a Hermes agent drive Blender, held back mainly by platform gaps and a bridge with no guard rails.

The plugin sits in Nous Research's hermes-official-plugins repo, and version 0.4.0, out on 24 October 2026, added the headline feature: Hermes installs Blender for you after one yes.

This verdict is built from the plugin's README, its workflow skill and Nous's own verification log, scored criterion by criterion.

The ratings are my editorial judgement from those docs, not a benchmark run.

If you'd like my team to tell you which agent tools are worth your time this quarter, book a free SEO strategy session.

The verdict in one line

Use it if you want Blender scenes read, edited and rendered from a chat, and skip it if your machine is an Intel Mac or your scenes can't risk unguarded Python.

What it is, and how it differs from the Blender MCP route

The plugin bundles two things: a launch config for the official Blender Lab MCP server and a Blender workflow skill.

The video above shows the earlier Blender MCP route, which meant wiring a server into Hermes yourself.

This review covers the official plugin, which replaces that wiring with one install and a skill that knows how to get Blender ready.

It replaces the archived NousResearch/hermes-plugin-blender, and it pins Blender Lab's source to one Git revision, add-on v1.0.3.

It is not the unrelated PyPI package also named blender-mcp, and the README tells you not to swap that in.

The scorecard

CriterionScore /10Why, from the docs
Setup effort9Two install commands, then one yes covers Blender plus the add-on, with no admin rights.
Scene control9execute_blender_code gives the agent full bpy access, plus scene summary, screenshot and render tools.
Install hygiene9sha256 checks on Blender and the add-on, user-folder installs, never over an existing Blender.
Transparency9A public verification log lists what passed, what failed first time and what hasn't run yet.
Live feedback8Desktop preview pane refreshes after each change; needs a graphical Blender.
Headless and server use7Background mode works for scene edits, but screenshots and viewport renders don't.
Safety defaults6Loopback only and asks before editing projects, but the port has no password and Python runs unguarded.
Platform coverage6No Blender 5.1+ for Intel Macs or Linux ARM; macOS and Windows installers not yet run on real machines.

Overall: 7.9 out of 10, the straight average of the eight scores.

Setup and scene control carry it, and those are the two criteria that decide whether people keep using a tool.

What Nous actually verified

Most plugin pages tell you it works; this one shows its working.

CheckSetupResult
Scene read, create, move, readback, deleteWindows ARM64 and x64, Blender 5.2.2 LTSPass on both
MCP tool listingSame machines26 tools
Agent opens Blender itself (0.3.0)Ubuntu x86_64, Blender 5.2.3 LTSScene answered in 52 seconds, no question asked
Agent installs Blender itself (0.4.0)Ubuntu x86_64, fresh home, no BlenderOne question, then a full install and scene read
Blender install on macOS and WindowsNot yet run on a real machineUnverified

The log even records its own slip: the first ARM64 attempt had a syntax error in the test's deletion code, the server returned that error, and the corrected run passed.

That kind of honesty is why transparency scores a 9.

It also notes the Windows runs were a protocol client smoke test, not a model working through the full Windows Hermes app.

Three strengths that earn the score

Strength 1: the one-yes install is the real story

Earlier versions needed you to have Blender ready.

Nous's log shows 0.3.0 refusing to install on a machine without Blender, with the message blender is not installed. Install blender 5.1 or newer.

On the same clean machine, 0.4.0 installed and enabled, asked one question on the first turn, and on "Yes, go ahead" pulled Blender 5.2.3, matched its hash, installed add-on v1.0.3 and read Camera, Cube and Light from the startup scene.

That's a full setup in two chat turns.

Strength 2: it opens Blender the right way

The skill launches Blender with --online-mode, which grants the add-on online access for that session only.

Your saved Allow Online Access preference stays as you set it.

The log backs the choice up: without the flag, the bridge never bound port 9876 in 40 seconds; with it, the bridge came up in 3.

It also refuses to start a second Blender window beside one that's already open.

๐Ÿ”ฅ Want tools like this tested and set up inside one agent system? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members comparing notes on what's worth running. โ†’ Join the AI Profit Boardroom

Strength 3: it checks its own work

The skill tells the agent to read the scene before editing, make small changes, and query the result after each one.

It also warns that a successful transport reply is not a successful scene change.

That habit, built into the instructions, is what separates a useful agent from one that reports success on a broken scene.

Plugin vs hand-wired MCP, side by side

JobOfficial pluginHand-wired MCP route
Server sourcePinned Blender Lab Git revisionWhatever you point it at
Blender installOffered after one yes, hash-checkedOn you
Add-on installPinned v1.0.3, covered by the same yesOn you
Opening BlenderAgent launches it with --online-modeYou open it and enable access
Workflow rulesSkill: read first, small edits, readbackOnly what you prompt

The hand-wired route can still work, as the video shows, but every row in the right-hand column is a step you can get wrong.

The plugin turns those steps into rules the agent follows each time.

Three limits that cost it points

Limit 1: no guard rails on the bridge

Blender Lab states that its tools run arbitrary Python in Blender without protection against file deletion.

Nothing protects the add-on's port with a password, which leaves it open to every local program for as long as Blender runs.

The README is blunt: this is not a sandbox, so use a disposable scene or a separate machine for experiments.

That's the main reason safety scores a 6.

Limit 2: platform gaps and unrun installers

Blender 5.1 and newer have no build for Intel Macs or Linux on ARM, so those users are out entirely.

The Linux install path ran end to end, but per the verification notes the macOS .dmg and Windows .zip install steps had not yet run on a real machine.

The Windows checks that did run were protocol smoke tests on ARM64 and x64 with Blender 5.2.2 LTS, listing 26 tools, not a full model run through the Windows Hermes app.

Expect the first Windows or Mac setup to need a little more patience.

Limit 3: small quirks that cost a retry

  • The first uv build of the server took several minutes in Nous's Windows checks, so the README asks you to warm it before connecting.
  • Readback code must set result to a dict with string keys, or it errors after the code already ran.
  • The _for_cli tools need blender on the backend's PATH, and BLENDER_PATH from a shell or .env doesn't reach the server.
  • A session that started without the Blender tools needs /reload-mcp or a new chat.

How it keeps your machine tidy

Installing the plugin changes nothing on your computer by itself.

Blender lands in a user folder, never in system folders, and never over an existing copy.

The skill unpacks into a scratch directory first and only moves the finished folder into place, so a failed step can't leave a half-installed Blender behind.

It deletes the download and leftover scratch files afterwards, even after a failure.

The only change to Blender's saved preferences is installing or enabling the add-on; online access comes from the launch flag, not a saved setting.

That's careful engineering, and it's why install hygiene earns a 9.

Who should use it

  • Hermes users on Apple silicon, Windows or x86 Linux who want Blender scenes changed from a chat.
  • Small studios and marketers who need quick scene edits and stills without a Blender specialist on every job.
  • Anyone already making models with the official Meshy plugin, which has a Blender handoff setting built in.

Who should skip it

  • Intel Mac and Linux ARM users, since Blender 5.1+ won't run there.
  • Teams that can't separate experiments from their only copy of a client project.
  • Anyone who needs screenshots or viewport renders from a headless server, which background mode doesn't cover.

๐Ÿ“ž Want a straight answer on which agent tools fit your business? Book a free SEO strategy session and my team will go through your marketing and the AI tools worth adding. โ†’ Book a free strategy session

How it rates against making the model first

This plugin is for driving Blender: scenes, edits, renders and scripts.

Generating a fresh model from text or a photo is a different job, handled by Meshy.

Our Hermes Meshy verdict and Hermes 3D review cover that side, and the two plugins work best as a pair.

Version history that shaped the score

VersionWhat changed, per Nous's log
0.2.0With Blender closed, the agent told the user to open Blender and enable the add-on.
0.3.0The agent started blender --online-mode itself and answered in 52 seconds with no question asked.
0.4.0The agent installs Blender and the add-on after one yes (24 October 2026).
0.4.1Current plugin.json version; the skill's search includes the Microsoft Store package on Windows.

What would push the score higher

Three changes would lift this plugin into the 9s.

A verified Blender install on real macOS and Windows machines would raise platform coverage.

An optional password or approval layer on the bridge would raise safety defaults.

And verified screenshot and liveness checks, which the README says the wrapper doesn't yet implement, would tighten live feedback.

Final verdict: 7.9 out of 10

It's the best route right now for a Hermes agent working inside Blender, and it beats a hand-wired MCP setup on install, launch and self-checking.

Fix your platform question first, keep experiments on scratch scenes, and it earns its place.

Check the Hermes Agent releases and Hermes docs for updates, and get the full agent stack with coaching inside the AI Profit Boardroom.

Also On Our Network

FAQ: Hermes Agent Blender Verdict

What score does the Hermes Blender plugin get?

7.9 out of 10 overall. Setup, scene control and transparency score 9, while safety defaults and platform coverage score 6.

Is the plugin better than the Blender MCP route?

Yes, for most people. It installs Blender and the add-on after one yes, launches Blender correctly and checks its own edits, which a hand-wired MCP setup leaves to you.

What is the biggest weakness?

The bridge has no guard rails. Blender Lab warns its tools run arbitrary Python, and the local port has no password.

Was it tested on Windows and Mac?

Windows ARM64 and x64 passed protocol smoke tests with Blender 5.2.2 LTS. The macOS and Windows Blender install steps had not yet run on a real machine.

๐Ÿ“บ Video notes + links to the tools ๐Ÿ‘‰

๐ŸŽฅ Learn how I make these videos ๐Ÿ‘‰

๐Ÿ†“ Get a FREE AI Course + Community + 1,000 AI Agents ๐Ÿ‘‰

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

  • I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
  • I've grown a YouTube channel to 400,000+ subscribers.
  • I wrote "Link Building Mastery".
  • My Udemy courses have taught over 29,000 students.

โ†’ Get my best AI training inside the AI Profit Boardroom

If your platform fits and you keep a scratch scene, install it now; for help choosing your next tools, book a free SEO strategy session or join the AI Profit Boardroom, because the verdict on Hermes Agent Blender is clear.