English · العربية · Español · Français · 日本語 · 한국어 · Tiếng Việt · 中文 (简体) · 中文（繁體） · Deutsch · Русский

Why This Exists

Creative tools are gaining agent bridges, but each bridge has a different install path, port, protocol, and safety model. AppAutoAction keeps those targets in one registry, validates them, emits MCP client config, and dispatches dry-run or live JSON envelopes to the right adapter.

It is intentionally small: Python standard library, explicit config, no hidden editor automation.

Quick Start

Install from npm:

npm install -g @lazyingart/app-auto-action
app-auto-action --version
app-auto-action webapp start --port 19473

Or run from a source checkout:

PYTHONPATH=src python -m agenticapp list
PYTHONPATH=src python -m agenticapp doctor
PYTHONPATH=src python -m agenticapp dispatch blender "Create a red cube at the origin" --dry-run
PYTHONPATH=src python -m agenticapp mcp-config
PYTHONPATH=src python -m agenticapp studio status
PYTHONPATH=src python -m unittest discover -s tests

After installation, the console command is also available as:

app-auto-action list
app-auto-action dispatch unity "Create a test scene with three labeled cubes" --dry-run
app-auto-action studio figure-grid "optical device icons 2x3" --rows 2 --cols 3
app-auto-action webapp start --port 19473

Paper Figure Studio

app-auto-action web --port 8787 --open

The web app now has a bright-by-default theme, chat panel, artifact canvas, scene editor, and backend settings. It can:

• Switch the visible studio UI across the same 11 languages as the localized READMEs.
• Treat generated overview images as concepts, then decompose them into editable atomic parts.
• Generate exact NxM SVG paper-figure grids with black panel boundaries.
• Prepare AgInTi image-generation dry-run payloads for scientific icon concepts.
• Store BioRender MCP settings without storing secrets.
• Export the current scene to OpenSCAD for mechanical layout planning.
• Render the scene through Blender and preview PNG, .blend, .scad, JSON, and text artifacts.
• Toggle Blender, OpenSCAD, AgInTi image generation, BioRender MCP, and target-registry routing settings.
• Dry-run any configured target from the studio and save the dispatch envelope as a canvas artifact.

Artifacts are tracked under output/webapp/artifacts.json and served in the canvas rail. The intended figure architecture is documented in docs/EDITABLE_FIGURE_PIPELINE.md. See also docs/PAPER_FIGURE_STUDIO.md, docs/STUDIO_CLI.md, docs/WEBAPP.md, and docs/NPM.md.

3D Experiment Design

AppAutoAction now includes a systematic Blender workflow for paper setup figures, optical benches, device concepts, and experiment design:

app-auto-action web --port 8787 --open
app-auto-action scene-template experiment-setup --output my-setup.scene.json
app-auto-action render-scene my-setup.scene.json --dry-run
app-auto-action render-scene my-setup.scene.json --output-dir output/scenes

The web app provides chat, JSON scene editing, dry-run planning, and render preview. The source of truth is a JSON scene spec. Blender runs headless and produces a .png preview plus a .blend scene. Start from examples/paper-optics-setup.scene.json, inspect the generated example render, or read docs/WEBAPP.md and docs/SCENE_SPEC.md.

Local Blender Test

For a no-sudo local Blender install and a real headless scene generation test:

scripts/install_blender_portable.sh
app-auto-action --config configs/blender-local-command.example.json doctor
app-auto-action --config configs/blender-local-command.example.json dispatch blender "Draw a welcoming modern building with a tower"

The command bridge is bridges/codex_exec_blender.sh. It reads the AppAutoAction JSON envelope from stdin, runs Blender in background mode, stores Blender logs under output/blender/, and returns clean JSON with .blend and .png artifact paths.

Targets

| Target | Current adapter | Best bridge shape | Notes | | --- | --- | --- | --- | | Blender | http_json | Blender MCP add-on, local HTTP, or command bridge | Good for scene generation, materials, rendering, export. | | AgInTi | local_command via web settings | aginti image --json | Dry-run image payloads for figure concepts; live calls require provider keys. | | BioRender | browser plus MCP metadata | Official remote MCP connector | Use OAuth/API-supported flows; avoid scraping. | | Unity | http_json | Unity package, WebSocket proxy, or C# editor bridge | Good for scenes, assets, scripts, tests, play mode. | | Unreal | http_json | Unreal MCP plugin or Python remote execution proxy | Treat as privileged editor access. |

Copy configs/targets.example.json to agenticapp.targets.json for local ports, commands, and tokens. This override file is ignored by git.

Research-Backed Design

The design follows the MCP split between tools, resources, and prompts, then adapts it to live editor bridges. The research brief is in docs/RESEARCH.md, covering:

• Blender MCP projects with headless and live-GUI modes.
• Unity MCP packages with scene, asset, script, and play-mode control.
• Unreal MCP servers using plugins or Python Remote Execution.
• BioRender's documented MCP connector endpoint.
• Security tradeoffs for agents with editor write access.

Architecture

Agent or MCP client
        |
        | command / dry-run / MCP config
        v
AppAutoAction CLI
        |
        | target registry
        v
Transport adapter: http_json | local_command | browser | noop
        |
        v
Blender / BioRender / Unity / Unreal bridge

Every dispatch receives the same envelope:

{
  "target": "blender",
  "kind": "blender",
  "instruction": "Create a red cube at the origin",
  "payload": {},
  "metadata": {
    "source": "agenticapp"
  }
}

Languages

Localized READMEs live under i18n/ and use the same profile-style language switcher as this root README:

العربية · Español · Français · 日本語 · 한국어 · Tiếng Việt · 中文 (简体) · 中文（繁體） · Deutsch · Русский

Development

PYTHONPATH=src python -m unittest discover -s tests
PYTHONPATH=src python -m agenticapp doctor

Keep transport behavior covered by tests before adding live editor features. See AGENTS.md for contributor guidance and SECURITY.md for the editor-automation security model.