tiobeharxi1987/blender-ai-copilot-flow

A controlled AI workflow for Blender: every copilot edit is a supervised tool call—proposed, approved, executed in sequence, cancellable after the current step and undoable by name—inside a visual studio with a built-in MCP server.
[download-url]: https://github.com/tiobeharxi1987/blender-ai-copilot-flow/releases
[ ]
[Download][download-url]
Nothing lands unseen: every edit is proposed, approved, run in order, and reversible by name.
## Why
The scary part of AI in a production scene is not what it builds—it is what it might change while you blink. blender-ai-copilot-flow makes every copilot action a visible step in a controlled flow: proposed before execution, approved by you, applied in sequence with earlier steps finished first, cancellable after the current operation, and undoable through named undo steps. Tool calls are validated against the allowed set before they run. You get the speed of an AI copilot with the audit trail of a pipeline.
## Without / With
| Without | With |
| Fear of a wrecked scene | Nothing runs without your approval |
| Mystery diffs you discover later | Each operation named and proposed before it executes |
| Parallel edits colliding mid-brief | Sequential execution—one ordered flow |
| No way back after a bad run | Named undo steps for every mutation |
| Runaway multi-step plans | Cancel halts the workflow after the current step |
| "Is this call even allowed?" | Python tool validation catches disallowed operations |
## Quick Start
Every command below is self-contained: open PowerShell, copy it, paste it, press Enter.
1. Download mcp-for-blender-pro.zip from Releases.
2. In Blender, open Edit → Preferences → Get Extensions → menu → Install from Disk.
3. Select the ZIP, enable mcp-for-blender-pro, then open 3D Viewport → N → MCP Pro.
4. Choose a provider, add your connection details and select a model.
5. Send a message in Blender or click Open Studio Window.
## Features
Approve every operation—each tool call is proposed before it runs.
Cancel safely—Stop halts the workflow after the current step, mid-plan.
Undo by name—mutations leave named undo steps you can pick from Blender's own controls.
Run edits in sequence—ordered steps, no collisions between operations.
Validate tool calls—disallowed operations are caught before execution.
Watch each step—viewport snapshots confirm what actually changed.
Serve the flow to MCP clients—external calls meet the same supervised pipeline.
## How it works
blender-ai-copilot-flow runs inside Blender as an extension: the copilot inspects your scene through 27 supervised tools, proposes each brief as a sequence of named operations, and executes them in order with your approval, a safe cancel and named undo per mutation. Python tool validation checks each call against the allowed set before it runs—it catches disallowed operations, though it is not a security sandbox—and the same supervised flow is exposed to external clients through a local authenticated endpoint (http://127.0.0.1:9876/mcp).
## Usage
A careful production pass. "Add the three lights and the camera, one step at a time—show me each before the next." Approve, watch, continue; cancel anywhere after the current step.