AZTechnology
한국어
BlenderBlender and AI

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.

AZTechnology15 min read

한국어English

Based on: Blender 5.2 LTS (screenshots from 5.2.1, English UI, default keymap), Blender Lab MCP add-on and server 1.0.3, Claude Code 2.1.318. Checked in November 2026. All figures are real Blender screens and renders. The ten prompts were sent one after another in a single conversation.

Once Blender MCP is connected, it is hard to know what to ask first. This guide is a record of ten first prompts that we actually sent. For each one, it shows which tool the AI used, how the Blender screen changed, and what a person should check.

The prompts go from easy to harder. The first three only read the scene. Then we create objects, edit them, render, and finally review the changes. Think of a first driving lesson: start the engine, read the dashboard, then drive slowly around the parking lot.

After this guide you will be able to:

  1. Check the connection with read-only prompts that do not change the scene. (Section 1)
  2. Send prompts that create and edit objects, then check the result with numbers and the screen. (Sections 2–3)
  3. Have the AI set up the camera and save a render to a file. (Section 4)
  4. Ask the AI to list its changes, and spot mistakes in its answer. (Section 5)

Before you start:

  1. Install the official MCP add-on in Blender and start its server. See Installing the official Blender MCP.
  2. Connect the MCP server to your AI app. For Claude Code, follow Adding the official Blender MCP to Claude Code. To pick another app, see Choosing an AI tool for Blender MCP.
  3. In Blender, open a new file with File > New > General. Every prompt here starts from this default scene (Cube, Camera, Light).
  4. Do not practice on a file you are working on. Code that the AI runs really changes the file. See Safety check before handing Blender to AI.

The ten prompts at a glance

No.PromptTypeChanges the scene
1Tell me which objects are in the open Blender scene. Don't change anything.ReadNo
2Tell me the location, size, and rotation of Cube.ReadNo
3Capture the Blender window and describe what is on screen.ReadNo
4Lay down a 10 m by 10 m floor plane. Name it Floor.CreateYes
5Put a monkey head (Suzanne) on top of Cube, touching its top face.CreateYes
6Make a red material and put it on Cube.EditYes
7Add a Subdivision Surface modifier with 2 levels to Suzanne and make it smooth with Shade Smooth.EditYes
8Aim the camera so it frames all the objects on Floor, and switch the 3D Viewport to the camera view.CameraYes
9Render one 960x540 image with EEVEE and save it as E:\mcp_demo\render_01.png.RenderCreates a file
10List what was changed in the scene during this conversation, and tell me what to delete or change to get back to the start.ReviewNo

For prompt 9, use a folder that exists on your PC. Create it first if needed.

1. Three read-only prompts: check the connection

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

Read-only prompts do not change the scene, so they are the safest way to check the connection. "Don't change anything" at the end of prompt 1 draws a clear line for the AI.

Steps:

  1. Send prompt 1 to your AI app.
  2. Check that the answer lists Camera, Cube, and Light.
  3. Send prompts 2 and 3 in turn.

The default Blender scene. A selected gray Cube in the middle of the 3D Viewport, Camera on the left, Light at the top right, the Outliner with three rows Camera, Cube, Light, and the Properties editor showing Location 0 m, Rotation 0°, Scale 1.000 Figure 1. The default scene before any prompt. Cube in the middle, Camera on the left, Light at the top right, three objects in the Outliner (Blender screenshot)

The tools the AI used and its answers:

No.Tool usedMain points of the answer
1Objects summary (get_objects_summary)3 objects (Camera, Cube, Light), all in Collection, Cube is the selected active object, same as the default scene
2One object in detail (get_object_detail_summary)Location 0, rotation 0, scale 1, a 2 m cube, one material named Material
3Window capture (get_screenshot_of_window_as_image)A description of each screen area (3D Viewport, Outliner, Properties editor, Timeline), 8 vertices and 6 faces in the status bar, version 5.2.1

All three answers ended with a line saying the AI only read the scene and changed nothing. Even when that line is there, get into the habit of checking the Outliner and Properties editor yourself.

In answer 2, Scale and Dimensions are different values. Scale 1 means "original size", and the default cube is 2 m on each side. A photo at 100% zoom still shows an object with its own real size.

When you ask for a screen capture, as in prompt 3, the AI sees the Blender window as an image. Answer 3 went through what is visible in Figure 1, area by area. When the number-based answers (1 and 2) agree with the image-based answer (3), the connection is working.

2. Two prompts that create: a floor and a monkey head

Starting point: continue from the scene in Section 1. Nothing has changed yet.

Steps:

  1. Send prompt 4.
  2. Check that Floor appears in the Outliner.
  3. Send prompt 5.
  4. Check in the 3D Viewport that the monkey head sits on the cube.

3D Viewport. A selected gray Cube on a wide gray floor plane, with a faceted monkey head resting on it by the tip of its chin and seeming to lean forward. The Outliner shows Camera, Cube, Floor, and the status bar shows Objects 1/5 Figure 2. The scene after prompts 4 and 5. Floor laid at the height of the cube's bottom, Suzanne touching the top face with the tip of its chin, five objects in the status bar (Blender screenshot)

For both prompts the AI created the objects with the code execution tool (execute_blender_code). Before running code, it read the scene units and checked whether an object with the same name already existed. Afterwards it captured the screen to check the result.

The AI made some decisions on its own. They were not in the prompt, so a person needs to check them.

PromptWhat the AI choseReason it gaveWhere to check
4Put the floor at Z = −1 mThe cube's bottom is at −1 m, so at 0 the cube would be half buriedFloor's Location Z in the Properties editor
5Matched Suzanne's lowest point (the chin tip) to the cube's top (Z = 1 m)Read "touching the top face" as the lowest point touching itOnly the chin tip touches, as in Figure 2
5Kept the default sizeSaid the distance between the ear tips is 2.73 m, wider than the 2 m cubeBoth ears stick out past the cube

When a word is open to more than one reading, such as "touching", the AI picks one. This time it made only the chin tip touch, and said so in its answer. If you wanted the head standing upright, you would need to describe the pose: "stand it upright facing forward, with the bottom of the head on the top face."

Answer 4 also asked back: "Tell me if z = 0 is better." When the AI tells you the values it chose, you can fix them in the next prompt.

3. Two prompts that edit: material and smoothing

Starting point: continue from the scene in Section 2 (with Floor and Suzanne).

Steps:

  1. Send prompt 6.
  2. Check that the cube looks red. Figure 3 is in Solid view; the AI also changed the viewport display color, so the cube looks red even in Solid view.
  3. Send prompt 7.
  4. Check that the faceted faces of the monkey head became round.

3D Viewport. A red Cube on the floor plane, with a smooth monkey head on top. The status bar shows Verts 7,970 and Faces 7,879 Figure 3. The scene after prompts 6 and 7. The red Cube, the rounded Suzanne, and much larger vertex and face counts in the status bar (Blender screenshot)

No.What the AI didWhat it told us
6Made a new material Red and put it in Cube's material slot 0. Set both Base Color and the viewport display color to red (0.8, 0, 0)A warning that the old material Material is no longer used by any object, so it will disappear after saving and reopening
7Added Subdivision Surface at 2 levels for viewport and render and applied Shade Smooth. Did not apply the modifierThe rounded chin tip had lifted about 1.26 cm, so it lowered Suzanne by that much (Z 1.984 → 1.972 m)

The warning in answer 6 is worth knowing. Blender drops data that nothing uses when you save. It is like throwing out unlabeled food when you clean the fridge. To keep the old material, turn on its Fake User (the shield button next to the name).

In answer 7, the AI did one more thing we did not ask for. Since we had asked for "touching the top face" earlier, it moved the head again so it still touched after the shape changed. That is handy, but it also means values you did not mention can change. When an answer says "moved" or "changed", check that value in the Properties editor.

What Subdivision Surface does is explained in Blender modifiers.

4. Camera and render: getting a file out

Starting point: continue from the scene in Section 3 (red Cube and smooth Suzanne).

Steps:

  1. Send prompt 8.
  2. Check that the text at the top left of the 3D Viewport changed to Camera Perspective.
  3. Check that Cube and Suzanne fit inside the dashed rectangle (the camera frame).
  4. Send prompt 9, with a folder on your PC.
  5. Open the saved PNG file in File Explorer.

The 3D Viewport in camera view. Camera Perspective at the top left, the red Cube and smooth monkey head inside the dashed camera frame in the middle, with the floor edges going outside the frame Figure 4. The camera view after prompt 8. Camera Perspective at the top left, Cube and Suzanne inside the dashed frame, floor edges outside it (Blender screenshot)

For prompt 8 the AI read "the objects on Floor" as Cube and Suzanne. It did not include the floor plane itself, so the floor edges go outside the frame. It kept the camera at its place looking down from the front right, aimed it at the middle of the two objects, and then moved it back until both fit. The changed values:

ItemBeforeAfter
Camera location (m)7.359, −6.926, 4.9588.507, −7.818, 5.254
Camera rotation63.6°, 0°, 46.7°68.4°, 0°, 46.7°
Lens, resolution50 mm, 1920×1080Unchanged

For prompt 9, the AI said it set the resolution to 960×540 only while rendering and put it back to 1920×1080 afterwards. It also checked whether a file with the same name already existed before saving. On my PC the render took about 0.9 seconds and the file was about 355 KB.

A 960x540 EEVEE render. A red cube on a gray floor, a smooth gray monkey head resting on it by the chin tip, and a long shadow falling back to the left Figure 5. The render file saved by prompt 9. The red Cube, Suzanne touching the top face with its chin tip, and the shadow from the default light (Blender render)

Answer 9 also said: "I could not look at the rendered image itself. Please open the file and check the composition." The AI only checked that the file was a 960×540 PNG. Whether the picture looks good is for a person to judge. In Figure 5 the monkey head rests on a single point at its chin and seems to tip forward. This is how the reading of "touching" from Section 2 shows up in the render.

5. The review prompt: undoing changes and the AI's mistake

Starting point: continue from the scene in Section 4.

Steps:

  1. Send prompt 10.
  2. Compare the list in the answer with the Outliner.

Answer 10 listed eight changes in a table: the new Floor, Suzanne, and material Red, Suzanne's modifier and position, the camera's location and rotation, the switch to camera view, the render result inside Blender, and the PNG file on disk. It also gave a way to undo each one and the camera's original values.

But the last lines held a wrong suggestion: "Since the file was never saved, use File > Revert to go back to the last saved state." File > Revert reopens the saved file. A new file that was never saved has nothing to go back to, so the menu item is grayed out.

Two lessons from this mistake:

  • An AI answer can sound right and still be wrong. Menu names and ways to undo are worth trying yourself before you trust them.
  • If you save before sending prompts, File > Revert really can take you back. Pressing Ctrl+S once before prompt 1 is the easiest undo you can set up.

6. Tips for writing prompts

Tips from sending these ten prompts:

TipExampleWhy
Use the object's exact name"on Cube", "to Suzanne"Matching the Outliner name keeps the AI from guessing
Add units"10 m by 10 m", "960x540"Without units the AI guesses them
Draw a line for read-only questions"Don't change anything"Keeps a read question from turning into an edit
Describe the pose"standing upright, facing forward"The AI picks one meaning for loose words like "touching"
Give the full save path"E:\mcp_demo\render_01.png"Without a path, files may land in a temp folder
Ask for one or two things at a timePrompts 4 and 5 sent separatelyEasier to check each result

More general advice is in How to write good prompts.

Practice

  1. In a new file, send prompt 1 and check that the number of objects in the answer matches the Outliner.
  2. Change prompt 5 to "Put Suzanne on top of Cube standing upright and facing forward, with the bottom of the head touching the top face." Note what is different from Figure 2.
  3. Save with Ctrl+S before prompt 1, then send prompts 4 to 7. Afterwards, use File > Revert and check that you are back at the starting scene.

Common mistakes and fixes

SymptomCauseFix
The first prompt says it cannot connect to BlenderBlender is closed or the MCP server stoppedGo through When Blender MCP won't connect in order
I only asked to read, but the scene changedThe prompt could also be read as a change requestAdd "Don't change anything". Undo with Ctrl+Z
An object appeared at the wrong heightNo reference height was givenSay "on the floor plane" or "at Z = 0"
The render file is nowhere to be foundNo save path was given, or the folder does not existCreate the folder first and give the full path
The original material vanishedMaterials used by no object are dropped on saveTurn on Fake User, or put it back in a slot
File > Revert is grayed outThe file was never savedSave with Ctrl+S before sending prompts

In one line

Three read prompts to check the connection → create and edit one or two things at a time → check every value the AI says it "changed" → save before you start. To use the same server from another AI app, compare the options in Choosing an AI tool for Blender MCP.

References

Back to the Blender guides

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.

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.

EEVEE render of a low-poly room with two walls and a wooden floor. A bookshelf with books on its shelves at front left, a bed with a blue blanket, a pinkish rug, a desk, chair, and desk lamp under the window, and a potted plant in the right corner (Blender render)Blender
Blender and AI13 min read

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.

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.