Skip to main content

A plated crab walker on a desert stage at sunset, built and rendered in Blender

MifBlender

Your AI assistant, working inside a running Blender.
428 typed tools for modeling, materials, UVs, rigging, animation, physics, geometry nodes and rendering.
Each one checks its parameters, does the work on Blender's main thread and reads the result back.

PyPI Free Blender 3.6 to 5.2 Windows

Start in one line

With uv installed and Blender closed:

uvx mifblender-mcp setup

It finds your Blender installs and AI apps, shows what it will do and asks first. Then it installs and enables the add-on, and adds MifBlender to the apps you pick. Restart your AI app, open Blender, and ask:

Call bl_status, then add a rounded cube with a brushed metal material and render a preview.

Free for any use, including commercial work. No account. Everything runs on your computer.

See what it can do

The crab walker's walk cycle

The crab walker: 515 pieces, an interior, a rig and a walk cycle, built with these tools in a long, guided session.

The form, in clay The topology
Clay render of the crab walker Wireframe close-up of a crab walker leg
The cabin Every angle
The crab walker's lit wooden cabin Eight views around the crab walker

The crab walker's game rig in wireframe through one walk cycle

From images to a model

The creature was built from one six-view character sheet in eight staged sessions, from blockout to a rigged walk. Each session started fresh, with only the saved file and a running journal to go on, and every round was checked against the sheet with bl_compare_to_reference.

The creature's walk cycle from a three-quarter camera and from the side

The finished model: 910,104 vertices with a painted tone on each vertex, and a 77-bone rig with IK on the arms and legs. In the walk loop, rendered in Cycles, the planted feet slide at most 1.9 mm per frame, the last frame matches the first exactly, and the claws stay within 0.008 mm of the fingers.

Reference and model The face
The creature from the front, beside its reference sheet Close-up of the creature's head beside its reference

The finished creature from eight angles

What the build taught:

  • A good score is not a good look. The outline matched the sheet at .88 to .90 while the body still read as a balloon. Numbers catch the shape, and someone still has to look at the picture.
  • Stop when the rounds stop helping. When the scores had been flat for half an hour, more rounds bought nothing, so the work moved to the next stage.
  • Reference views can disagree. On a painted sheet, the 3/4 views did not match the front, side and back, so the build held to those three.
  • Honest numbers can fall. Real fingers scored lower than the mitts they replaced, because the mitts had covered the painting by accident. Silhouette fit after the surface stage: front .85, side .89, back .84.
  • A dense mesh needs another way to weight. Blender's heat weighting failed on a mesh this size, so the weights were computed from capsules around each bone instead.

One prompt, one model

Each of these was built from the one prompt under it:

A boxy robot on tank treads
"Model a cute boxy robot with antenna, round eyes and tank treads, lit for a nice render."
A bonsai tree in a shallow pot
"Model a bonsai tree in a shallow ceramic pot with a few small rocks, lit for a nice render."
A low-poly cartoon car
"Model a low-poly cartoon car with round wheels, headlights and a little spoiler, lit for a nice render."
A teapot and cup on a wooden tray
"Model a cute ceramic teapot with a cup and saucer on a small wooden tray, lit for a nice render."
A stone castle tower with a flag
"Model a stone castle tower with battlements, arrow slits and a flag on top, lit for a nice render."
An open treasure chest spilling gold coins
"Model a wooden treasure chest with iron bands, its lid open, and gold coins spilling out, lit for a nice render."

Simulation, shading and scale

Scripted scenes, each built entirely from typed calls:

Manhattan as a scale model
Manhattan as a scale model
Liquid poured over a head
A baked liquid sim
125 toys poured into a crate
125 rigid bodies
A robot arm carrying a block
IK with a baked carry
27 cubies painted face by face
27 cubies, painted face by face
A spiral stair
A spiral stair from one tread
A geodesic lit from inside
A geodesic lit from inside
A furred sphere
30,000 hairs from one call
One render shown as beauty, normals and mist
Render passes

Game props

Clean to worn, from one setting Topology you can check
A table, crate, chair and barrel, clean and worn Five game props in clay and wireframe

What is in it

Area Examples
Modeling primitives, mesh editing, bevels, booleans, modular kit pieces, rocks and trees built by rule
Materials and UVs shader node graphs, texture sets, baking, unwrapping
Rigging and animation armatures, Rigify, poses, keyframes, walk cycles
Simulation rigid bodies, cloth, particles, geometry nodes
Rendering lights, cameras, Cycles and EEVEE stills and sequences, compositing
Checking mesh quality reports, comparing a render to a reference image, reading results back

A lean default sends only find_tools, mif_call and run_python up front and finds the rest on demand, so a long session stays small. Set MIF_TOOL_PACKS=blender to load all 428 tools, or name the packs you want; there are 13.

AI apps

setup knows Claude Code, Claude Desktop, Cursor, Windsurf, VS Code and Codex. For any other MCP client, start this as a stdio server:

uvx mifblender-mcp

Useful setup options: --dry-run shows the plan and changes nothing, --check reports what is set up, --upgrade moves the add-on and the apps to this version together, and --uninstall removes them. If several Blenders are found, it asks which.

Settings

Set these in the client's env block when you need them:

variable default
MIF_BLENDER_PORT 8792 the port the add-on listens on
MIF_BLENDER_TOKEN set by setup must match the add-on's Token preference
MIF_TOOL_PACKS lean lean, blender, or a list of packs
MIF_BLENDER_TIMEOUT 180 seconds one call may take

Safety

  • Everything stays on your computer. The add-on listens on 127.0.0.1 only, and setup generates a private token that links it to your AI apps. Neither makes any network call of its own.
  • run_python is on by default and runs any Python inside Blender. Untick Allow run_python in the add-on's preferences to limit the assistant to the typed tools.
  • The bl_gen_* tools are generative: they send a prompt to a ComfyUI you run yourself, and do nothing without one. Every other tool builds from explicit parameters. Set MIF_GENERATIVE=off, or untick Allow generative tools, to switch them off.

Tested on

Blender 3.6.23, 4.2.17, 4.4.0, 5.0.1 and 5.2.1, on Windows. macOS and Linux have not been tested yet; reports are welcome.

License

Free, source available. The add-on is MIT. The server is free for any use, including commercial work, and you may share it unchanged through the channels its license names; you may not sell it or publish changed copies. It is not an open source license. See LICENSE.txt. Some comments cite internal documents that are not published.

MifBridge

MifBlender is the Blender half of MifBridge, a paid product that adds Unreal Engine 5.3 to 5.8 and the tools that move models between the two. MifBlender stands on its own and needs neither.

Questions and bug reports: open an issue.

Metadata

Release files for mifblender-mcp 1.3.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for mifblender-mcp 1.3.1
File Size Uploaded Provenance Actions
mifblender_mcp-1.3.1.tar.gz 2.7 MB With provenance Details for mifblender_mcp-1.3.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for mifblender-mcp 1.3.1
File Interpreter ABI Platform Actions
mifblender_mcp-1.3.1-py3-none-any.whl Python 3 none any Details for mifblender_mcp-1.3.1-py3-none-any.whl

Total release size: 5.4 MB

Release files / mifblender_mcp-1.3.1.tar.gz

Download URL mifblender_mcp-1.3.1.tar.gz
Size 2.7 MB
Tags Source
SHA-256 checksum
How to use checksums
cdf84949273e65c9da26fd6c69164571231d2bf3c4942c810ac36f3f28ecc2c9
BLAKE2b-256 checksum
How to use checksums
e6cb6e056306dffac02917fa766966a39a90b6a87d603f027cb4a988167b3900
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 9, 2026.

Transparency log

Release files / mifblender_mcp-1.3.1-py3-none-any.whl

Download URL mifblender_mcp-1.3.1-py3-none-any.whl
Size 2.7 MB
Tags Python 3
SHA-256 checksum
How to use checksums
fa96a130b2be2fe7476808dd4db8aa2d0bccfda68757788dd1c6b4defe364732
BLAKE2b-256 checksum
How to use checksums
aefa577d53828900564ad1356885a9dcd8db006d69c32db7d0228aadb93627ed
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 9, 2026.

Transparency log

Security for mifblender-mcp 1.3.1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for mifblender_mcp-1.3.1.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 9, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

1.3.1 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page