Thaqif Kamarul’s Post

Using Claude to Talk to Blender via MCP I discovered something remarkable today. In a 35-minute vibe-build sesh, I managed to sketch a rocket engine inspired by the Raptor 3 while incorporating variations in the fuel nozzle and the top segment of the combustion chamber. I’ve never modelled anything significant in blender, aside from a 1 half-assed donut tutorial 😅 The initial complexity barrier felt like too much time to invest in. This was circa 2017 and didn't look back. Until today. Observations: 1. It’s not perfect: Occasionally, you’ll need to make manual edits to remove artifacts. 3.7 Sonnet sometimes require extended thinking to re-attempt missed instructions. 2. Large token sizes: Single prompts often require manual “continue” commands about three times to proceed. 3. Processing time: Generations can take a few minutes, but this gives you time to multitask (or watch in awe really...) Tips: 1. Reference specific component names in your prompts for better steering. 2. Start with broad instructions, then refine details for specific sections. 3. Don’t forget to save! 💀 The cool thing about building this way is its JIT (Just In Time) applied learning, so I'm feeling a more acquainted with where the components and panels are. The implications of MCP are profound. It feels like the inflection point is near. What a time to be alive! Ref: 1. https://lnkd.in/g46dA9xV 2. https://lnkd.in/g5jQynat 3. https://lnkd.in/gsV9Jy6N 4. Install guide: https://lnkd.in/gxEZR6AR