A new open-source project called aftr turns Adobe After Effects into a tool your AI coding agent can drive directly. Released as a Show HN on July 12, 2026 by developer Arman Luthra, aftr exposes roughly 90 After Effects commands through a Model Context Protocol server, so an agent like Claude Code, Cursor, or Antigravity can build compositions, animate layers, apply effects, and render finished video from plain-language instructions. The tag line is blunt: "Puppeteer for After Effects."
Tell the agent "Create a 1080p 5-second comp, add a title 'LAUNCH' with fire behind it," and it sees every After Effects operation as a callable tool. What separates aftr from the handful of other After Effects MCP servers is an autonomous render loop that renders a segment, reviews the frames, and self-corrects before assembling the final cut. This is the shift from AI as a scripting shortcut to AI as a motion-design collaborator.
What aftr Actually Does
aftr is a Node.js bridge between an MCP client and After Effects. A controller process runs an HTTP and WebSocket server on 127.0.0.1:8787. Inside After Effects, a signed CEP panel connects back over WebSocket and executes incoming JSON commands as ExtendScript, then returns structured JSON results. The MCP server sits in front of that bridge, so any MCP-aware agent gets the full After Effects vocabulary as tools.
That vocabulary spans about 90 commands grouped into project and comp control (createComp, setCompSettings, duplicateComp), layers (addSolid, addTextLayer, addCamera, addLight, addShape), animation (setKeyframe, setExpression, setEase), effects (addEffect, setEffectParam, listEffects), and render (importFootage, addToRenderQueue, render). On top of the primitives sit one-call VFX presets: fireEffect, smokeEffect, glitchEffect, neonGlow, and cinematicGrade, plus text animators like wordReveal, charScale, and blurFade. The project is MIT licensed.

How aftr Compares to Other After Effects MCP Servers
aftr is not the first bridge between AI and After Effects. Several MCP servers already ship remote control of compositions and layers. What is new is the depth of the command set and the autonomous, self-correcting render pipeline. Here is how the main open-source options line up.
| Project | Commands | One-call VFX presets | Self-correcting render loop | License |
|---|---|---|---|---|
| aftr | ~90 | Yes (fire, smoke, glitch, neon, grade) | Yes | MIT |
| ishu86/after-effects-mcp | 70+ | Templates (lower thirds, title cards) | No | Open source |
| Dakkshin/after-effects-mcp | Core set | No | No | Open source |
| TheLlamainator/after-effects-mcp | Core set | No | No | Open source |
The other servers are strong at scripted control and templated motion graphics. aftr aims higher: it wants to render, look at the output with FFmpeg, and revise on its own. If you only need an agent to drop in a lower third or set keyframes, the lighter servers are simpler. If you want an agent to produce and iterate on a full multi-segment sequence, aftr's orchestrator is the differentiator.
The Self-Correcting Render Loop
The headline feature is aftr's orchestrator pipeline. You describe a video as a spec of segments in JSON. For each segment the loop runs: spec to applySpec (realize the layers), a scoped aerender pass (render just that segment), visual review through FFmpeg (inspect the actual frames), a structured delta (decide what to change), and a re-render until it passes. Segments that pass are stitched with an FFmpeg concat.
Two design choices make this reliable. First, every segment is idempotent: re-running wipes and rebuilds its layers, so state never drifts across iterations. Second, rendering is non-blocking, with aerender streaming progress back to the agent instead of freezing it. For a creator, that means you can hand aftr a rough brief and let it grind toward a watchable result rather than stopping at the first render.

Set It Up: A Step-by-Step Workflow
aftr targets After Effects 2024, 2025, or 2026 (24.x through 26.x), Node.js 18 or later with v22 recommended, and FFmpeg on your PATH. It runs on both Windows and macOS. Here is the fast path to driving After Effects from Claude Code.
- Start the controller. With Node.js 18+ installed, run
npx aftr-studio controllerfor a no-install start, ornpm install -g aftr-studiothenaftr controller. The bridge comes up on127.0.0.1:8787. - Deploy the panel. From the repo root run
npm run deploy:panel. It builds the JSX bundle, signs a .zxp, and unpacks it into your per-user CEP extensions folder. - Open the panel in After Effects. Fully quit and relaunch After Effects, then open
Window > Extensions > aftr. The panel connects back to the bridge over WebSocket. - Register the MCP server. In Claude Code run
claude mcp add --transport http aftr http://127.0.0.1:8787/mcp. For Claude Desktop, add anmcpServersentry pointingAE_BRIDGE_URLat the same address. - Prompt a comp. Ask the agent to create a comp, add a title, and apply an effect. It calls tools like
createComp,addTextLayer, andfireEffect, then renders when you ask.
A headless simulator lets you test the command flow without After Effects open, which is useful for scripting and CI before you touch a licensed workstation.
What This Enables for Motion Designers
The practical win is turning a repeatable motion-graphics task into a prompt. Batch lower thirds for a 20-video series, generate ten title-card variants for A/B testing, or spin up a templated explainer where only the copy changes: these are the jobs that eat an afternoon in the timeline and that aftr can hand to an agent. Because the loop reviews its own frames, you get a draft that is at least watchable rather than a blind render you have to fix by hand.
aftr also slots into the broader pattern of agent-callable creative tools. The same MCP client that runs aftr can call Runway MCP for generative footage or a ComfyUI MCP for image and video passes, then bring the results into After Effects for compositing and finishing. That composability is where AI stops being a single tool and becomes a pipeline you orchestrate from one chat.

Limitations to Know Before You Rely on It
aftr is early: at launch the repository sits at a handful of stars and ten commits, so treat it as promising rather than production-hardened. There are concrete ExtendScript constraints too. The JSX bundle must be ASCII-encoded because After Effects mishandles BOM-less UTF-8, modal dialogs freeze the bridge so alert() calls are off-limits, and keystroke injection is best-effort on Windows due to foreground-lock rules. Third-party plugin support requires running an introspection pass to extract each effect's parameter tree. None of these are dealbreakers, but they shape what you can safely automate today.
Frequently Asked Questions
What is aftr?
aftr is an open-source, MIT-licensed MCP server that lets AI coding agents such as Claude Code, Cursor, and Antigravity control Adobe After Effects. It exposes roughly 90 After Effects commands, one-call VFX presets, and an autonomous render loop through the Model Context Protocol.
Which After Effects and Node versions does aftr need?
It supports After Effects 2024, 2025, and 2026 (versions 24.x, 25.x, and 26.x), requires Node.js 18 or later with v22 recommended, and needs FFmpeg on your PATH for visual review and final concatenation. Windows and macOS are both supported.
How is aftr different from other After Effects MCP servers?
Other servers like Dakkshin's and TheLlamainator's focus on scripted remote control, and ishu86's adds motion-graphics templates. aftr's differentiator is the larger command set plus an autonomous orchestrator that renders a segment, reviews the frames with FFmpeg, and self-corrects before assembling the video.
How do I connect aftr to Claude Code?
Start the controller with npx aftr-studio controller, deploy the CEP panel with npm run deploy:panel, open it in After Effects, then run claude mcp add --transport http aftr http://127.0.0.1:8787/mcp to register the MCP server.
Is aftr production-ready?
Not yet. It launched in July 2026 as an early Show HN project with a small commit history. It is worth testing for repeatable motion-graphics tasks, but keep a human in the loop and expect rough edges around ExtendScript encoding, modal dialogs, and Windows keystroke handling.
Does aftr cost anything?
aftr itself is free and open source under the MIT license. You still need a licensed, activated copy of Adobe After Effects to run it, plus whatever your MCP client (Claude Code, Cursor, or Antigravity) costs.