Unlisted

    How do I connect you to Blender for 3d modeling?

    Stats
    Prompts
    2
    Tool calls
    11
    Messages
    33
    Assets
    0
    Meta
    Author
    Mikkel Malmberg@mikker
    Created At
    August 12, 2026
    Source
    Codex
    Share

    Prompts

    Thought 0 times. Used 0 tool calls. +0 -0 [Expand]
    Skills instructions
    Redacted
    System prompt
    Redacted
    Collaboration mode
    Redacted
    App instructions
    Redacted
    System prompt
    Redacted
    System prompt
    Redacted
    System prompt
    Redacted
    System prompt
    Redacted
    Environment context
    Redacted

    How do I connect you to Blender for 3d modeling?

    I’m using the OpenAI/Codex documentation skill because this is about extending Codex with a local tool connection. I’ll check the current supported setup and then map it to Blender.
    Thought 0 times. Used 5 tool calls. +0 -0 [Expand]
    Tool: exec
    exec
const r = await tools.exec_command({cmd:"sed -n '1,240p' ~/.codex/skills/.system/openai-docs/SKILL.md","workdir":"~/dev/nitro_kit","yield_time_ms":10000,"max_output_tokens":20000});text(r.output)

    Output Redacted
    Tool: exec
    exec
const r = await tools.web__run({search_query:[{q:"site:developers.openai.com Codex MCP server configuration"}],response_length:"short"});text(JSON.stringify(r))

    Output Redacted
    Tool: exec
    exec
const r = await tools.web__run({search_query:[{q:"site:developers.openai.com/codex MCP configuration config.toml mcp_servers"}],response_length:"medium"});text(JSON.stringify(r))

    Output Redacted
    Tool: exec
    exec
const r = await tools.web__run({open:[{ref_id:"https://developers.openai.com/codex/mcp"}],response_length:"long"});text(JSON.stringify(r))

    Output Redacted
    Tool: exec
    exec
const r = await tools.exec_command({cmd:"~/.dotfiles/bin/bootledoot","workdir":"~/dev/nitro_kit","yield_time_ms":10000,"max_output_tokens":2000});text(r.output)

    Output Redacted

    The best connection is a Blender MCP server: a small bridge that exposes Blender operations as tools I can call.

    Typical setup:

    1. Install a Blender MCP add-on inside Blender.
    2. Start its local server from Blender.
    3. Register the accompanying MCP command with Codex:
    codex mcp add blender -- <command supplied by the Blender MCP project>

    For example, if its documentation specifies a Python package:

    codex mcp add blender -- uvx <blender-mcp-package>
    1. Restart the ChatGPT desktop app or Codex.
    2. Verify the connection:
    codex mcp list

    You can also configure it through ChatGPT desktop:

    Settings → MCP servers → Add server → STDIO

    Once connected, you could ask me things like:

    • “Create a beveled desk lamp and render it.”
    • “Make this model manifold and suitable for 3D printing.”
    • “Apply realistic materials and set up three-point lighting.”
    • “Build this scene procedurally so it remains editable.”
    • “Export the selected objects as GLB.”

    A good Blender bridge should expose operations for executing bpy scripts, inspecting the scene, capturing viewport/render images, manipulating objects and saving .blend files. Restrict it to local connections and save versions frequently—executing arbitrary bpy gives the bridge extensive access to Blender and your files.

    Codex supports both local STDIO and HTTP MCP servers, and the desktop app, CLI, and IDE extension share the same configuration. See the official MCP setup documentation.

    If you send me the Blender MCP project you’re considering, I can inspect its instructions and give you the exact installation commands.

    what's a good blender mcp?

    I’ll compare the actively maintained Blender MCP options by capabilities, installation friction, and repository health. The important distinction is whether you want direct scene editing, screenshot feedback, or a broader asset-generation workflow.
    Thought 0 times. Used 2 tool calls. +0 -0 [Expand]
    Tool: exec
    exec
const r = await tools.web__run({search_query:[{q:"Blender MCP server GitHub Blender 3D modeling MCP"},{"q":"site:github.com Blender MCP server bpy"},{"q":"Blender MCP addon model context protocol maintained"}],response_length:"long"});text(JSON.stringify(r))

    Output Redacted
    Tool: exec
    exec
const r = await tools.web__run({open:[{ref_id:"https://www.blender.org/lab/mcp-server/"},{ref_id:"turn3search5"},{ref_id:"turn3search1"},{ref_id:"turn3search0"}],response_length:"long"});text(JSON.stringify(r))

    Output Redacted

    There is now an official experimental Blender project, which changes the recommendation a bit. I’m checking whether it is usable today or still more of a lab preview, then I’ll give you one default choice rather than a long catalog.
    Thought 0 times. Used 4 tool calls. +0 -0 [Expand]
    Tool: exec
    exec
const r = await tools.exec_command({cmd:"~/.dotfiles/bin/web https://www.blender.org/lab/mcp-server/","workdir":"~/dev/nitro_kit","yield_time_ms":30000,"max_output_tokens":12000});text(r.output)

    Output Redacted
    Tool: exec
    exec
const r = await tools.exec_command({cmd:"curl -L --max-time 20 -sS https://www.blender.org/lab/mcp-server/ | sed -n '1,220p'","workdir":"~/dev/nitro_kit","yield_time_ms":30000,"max_output_tokens":12000});text(r.output)

    Output Redacted
    Tool: exec
    exec
const r = await tools.exec_command({cmd:"curl -sS https://www.blender.org/wp-json/wp/v2/pages/96198 | jq -r '.content.rendered' | sed -E 's/<[^>]+>/ /g; s/&nbsp;/ /g; s/&amp;/\\&/g' | tr -s ' ' | sed -n '1,180p'","workdir":"~/dev/nitro_kit","yield_time_ms":30000,"max_output_tokens":12000});text(r.output)

    Output Redacted
    Tool: exec
    exec
const r = await tools.exec_command({cmd:"~/.dotfiles/bin/bootledoot","workdir":"~/dev/nitro_kit","yield_time_ms":10000,"max_output_tokens":2000});text(r.output)

    Output Redacted

    My current recommendation is the official experimental Blender MCP server, provided you can run Blender 5.1 or newer.

    Why:

    • Maintained through Blender Lab
    • Designed around Blender’s Python API and documentation
    • Supports inspecting and modifying live scenes
    • Less dependency on an unknown third-party maintainer
    • Available as an MCP bundle for compatible clients

    The catch is significant: it executes model-generated Python without guards. Blender explicitly recommends using a VM or machine without sensitive data. See the Blender Lab MCP page.

    For an easier, more established community option, I’d choose MCPBlender/blender-mcp. It has:

    • Simple uvx blender-mcp installation
    • Viewport screenshots, so I can inspect results
    • Object, material, lighting, rendering, and scene operations
    • Poly Haven and Sketchfab integrations
    • A large user base and MIT license

    Connect that one to Codex with:

    brew install uv
codex mcp add blender -- uvx blender-mcp

    Then download its addon.py, install it through Blender’s add-on preferences, open the BlenderMCP sidebar, and click Connect. Its documentation warns that it can run arbitrary Python and enables telemetry by default; you can disable telemetry with:

    codex mcp add blender \
  --env DISABLE_TELEMETRY=true \
  -- uvx blender-mcp

    My short verdict:

    • Best trust/provenance: Blender’s official experimental MCP
    • Best ease and visual modeling workflow: MCPBlender
    • Best exhaustive specialist toolset: RFingAdam/mcp-blender, with 218 tools—but I wouldn’t start there because the much larger tool surface adds complexity.

    I’d start with MCPBlender today unless you already use Blender 5.1 and are comfortable testing Blender Lab software.