Ranked roundup · Blender
Best MCP Server for Blender
There is now an official MCP server for Blender, which is rare — most creative tools leave MCP entirely to the community. The Blender Foundation ships one through its Blender Lab program, at v1.0.0 since April 2026, focused on scene analysis and giving a model a natural-language route into Blender's Python API. The community got there first and went further: ahujasid's blender-mcp has 26,623 GitHub stars, supports Blender back to 3.0, and layers asset libraries and AI model generation on top of scene control. One thing this page cannot offer is a hosted or no-code path. Blender is desktop software, and we found no managed MCP platform — Composio, Zapier MCP, or Klavis — with a Blender integration, so everything here runs on your machine. All four servers execute model-generated Python inside your Blender session; that is the capability and the risk in one. Rankings come from verifiable criteria — official status, maintenance recency, adoption, auth model, documented coverage — never paid placement. Every repository was checked against the live GitHub or Gitea API on September 1, 2026.
Facts checked September 1, 2026 · ranked on verifiable criteria · placement is never for sale
Founder, Automation Squad ·
The ranking
- Community
Most users: prompt-assisted modeling and scene building in Claude Desktop or Cursor, with the broadest feature set and the largest community behind it
Object and scene creation, modification, and inspection; material application; arbitrary Python execution inside Blender; Poly Haven models, textures, and HDRIs; Sketchfab and Poly Pizza model downloads; AI-generated 3D models via Hyper3D Rodin and Hunyuan3D. Works with Claude Desktop, Cursor, VS Code, OpenCode, and Antigravity. Requires Blender 3.0+ and Python 3.10+
Local — a Blender add-on plus a server process run via `uvx blender-mcp`, communicating over a TCP socket (default localhost:9876) · None at the MCP layer (local socket, no accounts). Optional third-party API keys for Sketchfab, Poly Pizza, Hyper3D, and Hunyuan3D integrations · MIT · 26,623★
- Unofficial — a community project, not from the Blender Foundation
- The execute_blender_code tool runs arbitrary Python in Blender; the project's own README warns to always save your work before using it
- Telemetry is enabled by default (can be disabled in add-on preferences or via environment variable)
- Asset and AI-generation integrations require separate third-party API keys, some tied to paid services
- Only one instance of the MCP server should run at a time
- Official · Blender Foundation (Blender Lab)
Blender 5.1+ users who want a first-party, minimal server for scene analysis, file inspection, and understanding complex setups
Python code execution in the running Blender instance or a background Blender process; blend-file summaries (data-block counts, missing external files, linked-library trees, path and save status); scene analysis such as polygon-count outliers by screen size; data-block renaming; Geometry Nodes explanation and documentation. Documented with llama.cpp as the reference client; distributed as an .mcpb bundle or from source
Local — a Blender add-on plus a separate stdio MCP server process launched by your client, relaying to Blender over a TCP socket; deliberately minimal two-part design · None (local TCP between add-on and server); no accounts or API keys involved
- Requires Blender 5.1 or newer — LTS and older releases are not supported
- Young project: v1.0.0 shipped April 27, 2026, with a deliberately small toolset — no asset libraries or model generation
- Blender's own documentation warns it executes LLM-generated code with no guards and recommends running it in a VM or on a machine without sensitive data
- Hosted on projects.blender.org (Gitea), not GitHub, so star counts are not comparable to the community servers, and no license file is present at the repository root
- Community
Users who want structured, verifiable tool calls instead of an LLM writing raw Python into their scene
A curated tool API instead of raw Python generation: layered atomic, macro, and workflow tools; goal-first routing where sessions start by declaring what the model is building; deterministic verification through inspection, measurement, and assertion tools rather than trusting vision output. Add-on supports Blender 4.0+, tested on 5.0; Python 3.11+; works with any MCP-compatible client
Local — Blender add-on (installed from a release zip) plus a server process, with Docker examples for running the server · None at the MCP layer (local); configuration via environment variables rather than accounts or keys · Apache-2.0 · 55★
- Unofficial community project with small adoption — 55 GitHub stars as of September 2026
- Last pushed June 27, 2026 — a slower maintenance cadence than the top two entries
- More setup friction than the alternatives: add-on zip plus a server run with router and profile environment variables, typically via Docker
- Community
Fully offline workflows where nothing leaves your machine — no cloud LLM, no API account
Prompt-driven Blender scene control using open models served locally through Ollama, so the entire pipeline — model included — can run offline; also connectable from standard MCP clients
Local — Blender add-on plus a local MCP server, designed around locally hosted models · None (fully local); no cloud accounts required when paired with Ollama · Custom (Blender Open MCP License) · 112★
- Custom non-standard license ('Blender Open MCP License', classified by GitHub as Other) — review it before commercial use
- Last pushed April 21, 2026 — over four months without activity as of verification
- Unofficial community project with modest adoption (112 stars)
- Output quality depends heavily on which local model you run through Ollama
Blender MCP servers compared
| Server | Maintained by | Runs | Auth | License |
|---|---|---|---|---|
| Blender MCP (ahujasid) | Community | Local — a Blender add-on plus a server process run via `uvx blender-mcp`, communicating over a TCP socket (default localhost:9876) | None at the MCP layer (local socket, no accounts). Optional third-party API keys for Sketchfab, Poly Pizza, Hyper3D, and Hunyuan3D integrations | MIT |
| Blender MCP Server (official, Blender Lab) | Blender Foundation (Blender Lab) | Local — a Blender add-on plus a separate stdio MCP server process launched by your client, relaying to Blender over a TCP socket; deliberately minimal two-part design | None (local TCP between add-on and server); no accounts or API keys involved | — |
| Blender AI MCP (PatrykIti) | Community | Local — Blender add-on (installed from a release zip) plus a server process, with Docker examples for running the server | None at the MCP layer (local); configuration via environment variables rather than accounts or keys | Apache-2.0 |
| Blender Open MCP (dhakalnirajan) | Community | Local — Blender add-on plus a local MCP server, designed around locally hosted models | None (fully local); no cloud accounts required when paired with Ollama | Custom (Blender Open MCP License) |
How to choose
Pick ahujasid/blender-mcp if you want the widest coverage and the most proven option — 26.6k stars, Blender 3.0+, plus Poly Haven, Sketchfab, and AI-generated models. Pick the official Blender Lab server if you run Blender 5.1 or newer and want first-party maintenance for scene analysis and file inspection, accepting a deliberately smaller toolset. Pick blender-ai-mcp if raw code execution makes you nervous — it routes through curated tools with deterministic verification instead. Pick blender-open-mcp if you want everything local, including the model, via Ollama. There is no hosted option; all of these run on your machine.
How this ranking works
Every server here is real and was checked against its repository or vendor documentation on the date above — stars and licenses come from the GitHub API, not from memory. Order reflects verifiable criteria only: official support, maintenance recency, adoption, auth model, and documented capability coverage. We do not run paid placements, and we list every caveat we find, including the unflattering ones.
Common questions
- Is there an official MCP server for Blender?
- Yes. The Blender Foundation ships one through Blender Lab, hosted on projects.blender.org, with v1.0.0 released in April 2026. It requires Blender 5.1 or newer and is deliberately minimal — the community-built ahujasid/blender-mcp predates it and has far broader capabilities and adoption.
- What does setup and authentication actually look like?
- Every Blender MCP server is two pieces: an add-on installed inside Blender and a local server process your MCP client launches, talking over a TCP socket (ahujasid's defaults to localhost:9876). There is no OAuth and no account — the only credentials involved are optional third-party API keys for asset integrations like Sketchfab or Hyper3D. The real security question is code execution: these servers run model-generated Python inside Blender, and the Blender Foundation itself recommends a VM or a machine without sensitive data.
- Can I use a hosted or no-code platform like Zapier MCP or Composio instead?
- No. Blender runs on your desktop, and we found no managed MCP platform offering a Blender integration as of September 2026. Every option on this page requires installing an add-on and running a local server process.
- Which Blender versions do these servers support?
- ahujasid/blender-mcp works on Blender 3.0 and newer, the widest range here. The official Blender Lab server requires 5.1+, which rules out LTS releases. blender-ai-mcp's add-on supports 4.0+ and is tested on 5.0.
MCP servers we build
Automation Squad publishes its own open-source MCP servers and agent skills, documented with the same rules this ranking uses.