Skip to content
Prism

How it works

The MCP server holds the logic, the panel runs each instruction, and your AI reads, reasons, and writes native keyframes.

Prism connects your AI to your live After Effects project so it can read what is really there, plan a change, and write the same native, editable layers and keyframes you would make by hand. Underneath, two pieces do the work: the MCP server that holds all the logic, and a small panel inside After Effects that runs each instruction. This page walks through how that fits together, end to end.

The short version
The MCP server is the product — it carries every tool and all the logic.
The After Effects panel is small and simple; it runs each instruction and returns the result, nothing more.
Your AI follows one loop: read the live comp, reason about it, write native keyframes.
Your project files stay on your machine; only the specific action you request round-trips to the server.

For why a live server beats a frozen plugin in the first place, see What is an MCP server for After Effects.

Two pieces: server and panel

Prism splits cleanly into a remote server and a local panel.

The MCP server runs Prism's 200+ tools and all of their logic. When your AI calls a tool, Prism generates the exact instruction for that action — baking in details like which comp or layer is in focus — and sends it to the panel on your machine.

The panel lives inside After Effects (Window → Extensions → AE Prism). It holds no logic of its own. It receives the instruction, runs it in After Effects, and returns the result. That is its whole job.

The MCP server is the product; the panel inside After Effects simply runs what it's told.

Why logic lives on the server

Keeping every tool server-side has concrete benefits. Tools improve, get fixed, and expand on the server — so you never reinstall the panel and never run a version that drifts out of sync with the rest. It also means the panel stays small and predictable: one executor, many tools.

The read → reason → write loop

Every Prism action follows the same three-step loop.

01
Read
live comp
02
Reason
plan the change
03
Write
native keyframes
One request, one loop

First the AI reads your active composition — layers, properties, keyframes, and markers as true state, not a description you typed. Then it reasons about that state and plans the change. Finally it writes the result as native, editable After Effects objects: the same keyframes and layers you would create yourself.

Nothing about this is destructive by surprise. Everything is undoable in a single step, and a restore point precedes anything that could remove or overwrite your work.

What the tools cover

The work is grouped into server-side tool categories — animation, composition, effects and presets, expressions, layers, media and assets, project management, properties and transform, timeline, and utility — plus tools that bring formats AI already understands into After Effects. There are 200+ tools and capabilities in all.

These surface through the Prism panel's three tabs — MCP, JSON, and Tools — starting with the MCP tab.

Privacy by design

Because the logic stays on the server and execution stays on your machine, your project files stay on your machine; only the specific action you request round-trips to the server. For exactly what data is handled and how, see the security page.

Where to go next

These pages run in order — follow them top to bottom to get from nothing to your first edit.

On this page