Control-Anything

Analyze anything, control anything — strongest unofficial way to hand any software surface to an LLM. Define once, ship as Python API / CLI / MCP server / REST / Web UI / OpenCLI spec — pick which channels to enable per deployment.

Rename history: api-anything 0.1.0-0.3.2 → invoco 0.1.0-0.3.1 → control-anything 0.1.0+ (this series, 2026-04-20). Semantic intent: reflect the real product — controlling anything via LLM, not just "call" (invoco) or "REST API" (api-anything). See CHANGELOG.md full rename note.

The idea: SSMCE

Single Source, Multi-Channel Emission. You write the function once; control-anything is the upstream orchestrator that hands it to Typer, FastAPI, FastMCP, Gradio, and OpenCLI spec emitters so every consumer (humans, agents, remote systems, beginners) gets the shape they need.

from control_anything import expose

@expose(channels=["api", "cli", "mcp", "ui"])
def blender_add_cube(x: float = 0, y: float = 0, z: float = 0) -> str:
    """Add a cube to the scene."""
    import bpy  # Blender official Python API (Tier 1 upstream)
    bpy.ops.mesh.primitive_cube_add(location=(x, y, z))
    return f"cube@({x},{y},{z})"

The same function becomes:

0.1.0 (first release under the new name) ships the full SSMCE surface: expose() decorator, registry, ChannelArtifact + build_channel() + emit(), and control-anything serve --channels a,b,c parallel serve — all preserved feature-complete from the previous invoco 0.3.1.

Install

pip install control-anything

No mandatory dependencies. Add channel support as needed:

pip install 'control-anything[cli]'     # Typer
pip install 'control-anything[mcp]'     # FastMCP
pip install 'control-anything[rest]'    # FastAPI + Uvicorn
pip install 'control-anything[ui]'      # Gradio
pip install 'control-anything[all]'     # everything

0.1.0 quick tour

invoco --version
invoco info --json
invoco list       # lists exposed functions in the current interpreter

from invoco import expose, registry

@expose
def greet(name: str) -> str:
    """Say hi."""
    return f"hello {name}"

print(registry()["greet"].channels)  # ['api', 'cli'] — default channels

Positioning

• Upstream ecosystems we stand on, not compete with: CLI-Anything (HKUDS), OpenCLI spec, Python bindings (bpy, cv2, qgis), vendor SDKs (anthropic, openai, boto3).
• Our layer: tier detection + routing + multi-channel emit. Invoco decides how a function reaches every audience, not whether the audience exists.
• Two audiences, one source: power users import / agents call MCP or CLI / beginners open the auto-generated UI. Nobody writes the function twice.

Roadmap

• 0.1.0 — expose() decorator + registry + CLI skeleton. ✅ shipped.
• 0.2 — channel emitters: Typer CLI, FastMCP, FastAPI, Gradio UI.
• 0.3 — OpenCLI spec emit + AGENT.md generation + tier-detection port from upstream api-anything lineage.
• 0.4 — CLI-Anything bridge (wrap HKUDS generated CLIs as typed Python APIs).
• 0.5+ — UIA generator for closed-source Windows apps (Tier 3A exclusive).

History

Invoco 0.1.0 supersedes the api-anything preview line (last published 0.3.1, will be deprecated once the channel emitters ship). Core tier-detection and UIA code from api-anything will be rewired under Invoco's SSMCE model in 0.3+.

License

Apache-2.0.