BlenderBlender and AI

Building a low-poly room with Blender MCP: the order of questions and what to fix by hand

Send five questions in one conversation through the official Blender MCP to build a low-poly room with walls, a window, furniture, props, and lights, then use the AI's notes, a check that does not change the scene, and the first render to fix the bookshelf direction, the default light left on, and a dim desk lamp by hand.

AZTechnology13 min read

한국어English

Based on: Blender 5.2 LTS (screenshots from 5.2.1, English interface, default shortcuts), the EEVEE render engine, the Blender Lab official MCP server and add-on 1.0.3, and Claude Code as the AI app. Checked in November 2026. All images are real Blender screens and real renders.

If you ask AI to "make a room" in one go, everything to check lands on you at once. Split the job into five questions, and you can look at the screen and correct course at each step. This guide records building a low-poly room with two walls by sending five questions in one conversation through the official MCP. It covers the order of the questions, the screen and the AI's answer at each step, the problems seen in the first render, and the values fixed by hand.

Low-poly is a style that uses few faces to make things simple, like toy blocks. Furniture made from a few boxes and posts does not look out of place, so it is a good job for AI. If the term is new, see What is the low-poly style? first.

When you finish this guide, you will be able to do the following.

  1. Split one room into five questions for AI.
  2. Decide what to check at each step.
  3. Ask AI for a "check" that does not change the scene.
  4. Use the first render and the AI's notes to pick what to fix, and fix it with panel values.

Before you start

You need: an AI app connected to the official MCP, and Blender 5.1 or later with the server running. If you have not connected yet, see Connecting the official Blender MCP to Claude Code. To pick an AI app, see Choosing an AI tool for Blender MCP.

  1. Open a new file with File > New > General.
  2. Put the 3D Viewport in Solid view. You check each step in this view.
  3. Send the five questions in turn in one conversation. The AI needs to remember the names and positions from earlier questions for the next one to follow on.
  4. Save with Ctrl+S after each question. If the next question goes wrong, you can go back to the saved file.

Why split the questions: if you ask for everything at once, it is hard to find where it went wrong. It is like checking a house in order: foundation, walls, windows, furniture.

1. Question 1: floor and walls

Starting point: the default cube in a new file (File > New > General).

Do this: send this to your AI app.

Delete the default cube and make only the floor and two walls, back and left, of a room 4 m wide, 4 m deep, and 2.6 m high. Walls 0.1 m thick, flat single-color materials for a low-poly look. Put the objects in a Room collection.

Making only two walls lets the camera look into the room. It is open at the front and right, like a dollhouse.

The viewport in Solid view. Two greyish walls stand at the back and left on a wood-colored floor Figure 1. The result of question 1: one floor and two walls (Blender screenshot)

The AI made the floor (Floor), back wall (Wall_Back), and left wall (Wall_Left) as one box each. Its answer said it put the 0.1 m wall thickness on the outside of the room so the inside is exactly 4 × 4 m. That makes the floor and back wall 4.1 m long on one side.

What to check: see that a Room collection appears in the Outliner with Floor, Wall_Back, and Wall_Left inside. Also check that the bottom of each wall meets the top of the floor (height 0).

2. Question 2: the window

Starting point: continue from the scene in section 1.

Do this: send this.

Cut a window hole in the back wall (1.2 m wide, 1 m high, 1 m above the floor) and add a window frame.

The viewport with a square window hole cut in the back wall and a white frame around it Figure 2. The result of question 2: a hole and a frame in the back wall (Blender screenshot)

There are two ways to cut a hole. One is a Boolean modifier, which carves the wall in the shape of a box, like a cookie cutter pressed into dough. The other is to rebuild the wall's faces around the hole. This time the AI chose the second way. Its answer said it "rebuilt the wall mesh with quad faces instead of a Boolean." The frame (Window_Frame) is four 6 cm bars joined into one object.

The AI added one more note. 1.2 × 1.0 m is the size of the hole in the wall; because the frame sits inside it, the open part you see is about 1.08 × 0.88 m.

What to check: select the back wall and open the Modifiers tab (wrench icon). If it is empty, as here, the hole is part of the wall's shape. If there is a Boolean, do not delete the cutter box in its Object field. Deleting the cutter removes the hole too.

3. Question 3: large furniture

Starting point: continue from the scene in section 2.

Do this: send this.

Put a bed (1 x 2 m) against the left wall, a desk and chair under the window, and a bookshelf at the front right. Use only a few boxes for each.

The viewport with a bed against the left wall, a desk and dark grey chair under the window, and a large box-shaped bookshelf at the front right corner Figure 3. The result of question 3: bed, desk, chair, and bookshelf made of boxes (Blender screenshot)

The AI joined several boxes into one object for each piece of furniture. The bed (Bed) has five: base, mattress, blanket, pillow, and headboard. The bookshelf (Bookshelf) has eight: sides, back, and shelves. The bookshelf sits at the front right corner of the room (X 1.85, Y −1.6), with its shelves facing into the room. Figure 3 looks straight at that corner, so the bookshelf seems to be at the front middle.

What to check: check with the position numbers too that each piece went where you asked. Select the bookshelf and read Location in the Object tab. Also check that the chair back is on the side away from the desk. If the back faces the desk, the chair is turned the wrong way. In this result it is right.

4. Question 4: props, lights, and camera

Starting point: continue from the scene in section 3.

Do this: send this.

Add a rug, a desk lamp, a potted plant, and a few books, plus sunlight coming in through the window and a room light. Point the camera into the room from the front right, up high.

The viewport with a rug, a lamp and books on the desk, and a plant in the back right corner added, an area light mark above the middle of the room, and the camera outline outside to the right Figure 4. The result of question 4: props, lights, and the camera (Blender screenshot)

The AI added the following.

WhatWhat the AI made
Propsrug (Rug) 1.4 × 1.0 m, desk lamp (Desk_Lamp, with a softly glowing shade), plant (Plant), books (Books): 16 on the shelf + 2 on the desk
SunlightSun light (Sun_Window), strength 4, shining from beyond the back wall
Room lightArea light (Ceiling_Light) 120 W at ceiling height 2.5 m
Lamp lightPoint light (Desk_Lamp_Light) 15 W, just under the shade
Camerathe existing Camera moved to (4.6, −2.4, 4.4), 19 mm lens

The end of the AI's answer had three notes.

  1. The Point light that came with the new file (Light, 1000 W) is still on, so the room may look flatly bright.
  2. From this camera angle, the back of the bookshelf faces the camera, so the books inside cannot be seen.
  3. There is no ceiling, so it did not make a lamp shade object for the room light.

Do this: check the camera view with numpad 0, then press F12 for the first render.

The first render. At front left the back of the bookshelf stands as a big brown box and hides part of the rug; the bed, desk, chair, and plant are visible, the window shows dark grey, and the lamp light is barely visible Figure 5. The first render. The back of the bookshelf blocks the front, and the lamp light is barely visible (Blender render)

The render matches the AI's notes. Only the back of the bookshelf shows, and it hides part of the rug. The room is bright but flat, because one default light shines evenly from above, and the lamp light is barely visible. Also, the render resolution was still the new file's default (1920 × 1080). If the question does not mention resolution, the AI does not change it either.

5. Question 5: a check that does not change the scene

Starting point: continue from the scene in section 4.

Besides obvious problems, objects may cut into each other or float. These are hard to find on screen. Ask AI for a read-only check, and it finds them as numbers.

Do this: send this.

Give me a table of the bottom height of every object and the pairs whose bounding boxes overlap. Do not change the scene.

Always include "do not change the scene." If the AI fixes things on its own during a check, it is hard to know what changed.

The AI's first line said it "only read the scene and changed nothing." The result is summed up below. Everything standing on the floor started at height 0, and the lamp on the desk and the books on the shelves sat exactly on their surfaces. Five pairs had overlapping bounding boxes.

Overlapping pairAI's judgment
back wall and window frameno cutting in. The wall's bounding box includes the window hole, so they look like they overlap
books and bookshelfonly touching. The books rest on the shelves
books and deskonly touching. Two books lie flat on the desktop
books and plantapart. The 18 books are one object, so their bounding box is wide enough to cross the room
chair and deskapart. The chair seat goes 9 cm under the desk, but below the top and between the legs

The AI did not stop at comparing bounding boxes. It split each object into its box pieces, measured again, and concluded that no object actually cuts into another. A bounding-box check is looser than the real shape, so when it says "overlap," look again on screen before deciding.

6. What to fix by hand

Starting point: continue from the scene in section 5.

Based on the first render (Figure 5) and the AI's notes, fix five things with panel values.

Do this:

  1. Default light left on: in the Outliner, click the camera icon on the Light row to turn it off for rendering.
  2. Bookshelf showing its back: select Bookshelf and set Rotation Z to 180 in the Object tab. The bookshelf turns in place so its shelves face the camera. Leave the books (Books) as they are. The bookshelf turns around its own center, so the books stay inside the shelves.
  3. Dim desk lamp: select Desk_Lamp_Light and raise Power in the Light tab from 15 to 60.
  4. Room now too dark: turning off the default light in step 1 makes the room clearly darker. In the World tab, raise Color to R 0.15, G 0.17, B 0.22 (a light blue-grey).
  5. Resolution: set the resolution to 960 × 540 in the Output tab.

The render after the fixes. The shelves of the bookshelf at front left show the books, the bed, rug, desk, chair, lamp, and plant are visible, and the background and window show a light blue-grey Figure 6. The render after fixing five things (Blender render)

The shelves and books are visible, and with the even default light gone, the sunlight patch on the rug and the plant's shadow are clearer. The background and the view through the window turn the light blue-grey of the World color. The lamp only brightens the desk under its shade a little; for a clearer glow, raise Power more or hand it to AI in the next question.

7. Writing better questions next time

What came up this timeWhat to put in the question in advance (example)
The back of the bookshelf faces the camera"Make the open side of furniture face the camera"
The default light was left on"Turn off the Light that came with the new file for rendering"
The lamp light is weak"Set the lamp strength so its light shows in the render"
The resolution stayed at the default"Render resolution 960×540"
18 books are one object"Group the books on the shelf and on the desk separately"

Rather than putting all of this in one question, splitting the steps as in this guide and adding a line or two at each step makes the result easier to check. And always read the last paragraph of the answer. This time the AI pointed out two of the things to fix first.

Practice

  1. Select the chair (Chair) and then the desk (Desk), read Location Y in the Object tab, and write down whether the chair is in front of the desk (smaller Y) or behind it.
  2. Send the check question from section 5 again and see whether it still reports "no object actually cuts into another" after turning the bookshelf.
  3. Ask AI to "hang a picture frame on the right wall." See how the AI handles the fact that there is no right wall.
  4. Save this room file. With the method in Turntable renders with AI, you can spin around the room.

Common mistakes and fixes

SymptomCauseFix
The window hole disappearedWith a Boolean, the cutter box was deletedUndo with Ctrl+Z or go back to the saved file
Only the back of the furniture showsThe furniture faces into the room but the camera looks from outsideTurn the furniture's Rotation Z or move the camera
The render is flatly brightThe new file's default light is onTurn Light off for rendering and set brightness with the World color
The render is too darkAfter turning off the default light, there is not enough light to fill the roomRaise the World Color and Sun strength a little at a time
The check says they overlap but the screen looks fineBounding boxes are looser than the shapeLook closely on screen, or ask again "do they actually cut into each other"
The AI changed the scene during a check"Do not change" was missing from the questionGo back to the saved file and send the question again

Summary

Floor and walls → window → furniture → props, lights, and camera → a "do not change" check → fix with numbers after reading the AI's notes and the render. In the next guide, Turntable renders with AI, you turn the finished scene and take many images.

Further reading

References

KEEP EXPLORING

Read next

EEVEE render of a grey fighter-jet model sitting at an angle on a beige floor, with a shadow to the lower rightBlender
Blender and AI11 min read

Cleaning up an AI-generated 3D model in Blender: merge split points, set the size, reduce faces

Import a fighter-jet model GLB made with Meshy into Blender 5.2, read its state from the statistics, merge split points with Merge by Distance, tidy normals and smooth shading, set the size, and cut the face count to one tenth with Decimate.

A PowerShell window. Running blender-mcp through uv with --transport http --port 9191 prints Started server process, StreamableHTTP session manager started, and Uvicorn running on http://127.0.0.1:9191 (terminal screenshot)Blender
Blender and AI11 min read

Using Blender MCP with ChatGPT: custom MCP servers in ChatGPT on the web, secure tunnels, and how Codex differs

What to know when you want to use Blender MCP with GPT: why the official Blender MCP cannot plug straight into ChatGPT on the web, the two routes OpenAI's docs describe (a public HTTPS address or Secure MCP Tunnel), and how they compare with Codex, the easiest way to drive Blender with a GPT model.

A 960x540 EEVEE render. A red cube on a gray floor with a smooth gray monkey head resting on it by the tip of its chin, and a shadow falling back to the left (Blender render)Blender
Blender and AI15 min read

Ten first prompts for Blender MCP: from reading the scene to rendering, seen through the real answers

After connecting the official Blender MCP to an AI app, we sent ten first prompts in order: read, create, edit, render, and review. This guide shows what the AI did for each one and what a person still needs to check, with screenshots.

EEVEE render of a bright gold metal monkey head on a grey floor, with sharp reflections on the brows, eyeballs, and ears and an interior photo background with beige wallpaper and a dark window (Blender render)Blender
Blender and AI14 min read

Letting AI set up lights and materials: three-point lighting and PBR over MCP, compared in renders

Through the official Blender MCP, hand the three-point lighting and the gold metal and matte floor materials of a monkey head scene to AI, read the real result in the panels, and compare three EEVEE renders (default light, the AI's result, and a hand-tuned version) to learn why metal looks dull and which values are worth changing by hand.

Front view of twelve fence pickets selected with an orange outline, standing at even spacing on the red X axis, with Count 6 in the Array panel of the Modifiers tab on the right (Blender screenshot)Blender
Blender and AI14 min read

Letting AI build a modifier stack: Array, Mirror, and Bevel over MCP, then fixing it by hand

Through the official Blender MCP, ask AI to stack Array, Mirror, and Bevel modifiers on one fence picket, read the real result in the Modifiers tab, and fix the picket count the AI read differently with one number. Two common mistakes, unapplied scale and the Mirror origin, are covered too.

Two bar charts. Left: tokens for nine runs (A1 to C3), mostly input read from cache. Right: estimated cost, highest for the first run of each group and lowest for C2 and C3 at 0.065 dollars (chart)Blender
Blender and AI12 min read

Measuring Blender MCP token use: the same job three times each, and how to save

With Claude Code and the official Blender MCP, we ran 'place five cylinders' in three question styles, three times each, nine runs in all. We measured input and output tokens, time, and estimated cost, and compared with real numbers how much it helps to skip screenshot checks and to give exact values.