After Effects MCP

A production-minded Model Context Protocol server for Adobe After Effects.

Headless Startup bridge · sync tool calls · hybrid high-level tools + JSX escape hatch · composition frame capture as images · resources & prompts for agents.

Build motion graphics, review cuts, and render YouTube-ready projects from an MCP client while keeping After Effects as the execution environment.

Documentation

• Installation guide
• Usage guide
• Changelog
• Join the Discord

Features

• Headless by default — no “open panel + Auto-run” required
• Sync-feeling RPC — tools wait for AE (request IDs, timeouts, clear errors)
• Motion tools — comps, layers (text/shape/solid/null/camera…), keyframes, expressions, effects
• Vision — ae_capture_frame returns a PNG still so agents can verify work
• Resources — ae://project, ae://docs/*
• Prompts — lower-third, kinetic type, debug comp
• Companion skill — skills/after-effects-mcp/SKILL.md
• Editorial planning — sentence-aware cut scoring, explainable EDLs, clean/retention/story modes, YouTube structure, chapters, and b-roll/graphic suggestions
• Quality workflow — multi-frame QA, saved preview reviews, approve/reject decisions, and autonomous or approval-gated builds
• Motion system — reusable brand-aware hook, lower-third, chapter, statistic, quote, subscribe, and end-screen templates
• One-shot motion videos — canonical blueprints, visual grammar, coherence validation, deterministic scene builds, multi-frame QA, repair passes, and approval-aware previews
• Export presets — automatic MP4, MOV, and PNG render presets with explicit output paths

Requirements

• Adobe After Effects 2022+
• Node.js 20+
• AE preference: Edit → Preferences → Scripting & Expressions → Allow Scripts to Write Files and Access Network

Connection model (important)

You should not need to restart After Effects every time you reconnect the AI.

How it works: when the MCP server starts (AI client launch), it runs AfterFX.exe -r ae-mcp-ensure.jsx against the already open AE session, restarts the headless poller, and a watchdog re-kicks if heartbeats die.

Install

git clone https://github.com/HeroicSwan/after-effects-mcp.git ae-mcp
cd ae-mcp
npm install
npm run build
npm run install-bridge

If AE was already open during install, restart AE once so Startup bootstrap loads. After that, only restart the AI client.

npm run health
# or force reconnect without touching AE:
node dist/index.js ensure

MCP client config

Claude / Cursor / generic

{
  "mcpServers": {
    "ae-mcp": {
      "command": "node",
      "args": ["C:/Users/YOU/Projects/ae-mcp/dist/index.js", "serve"]
    }
  }
}

Grok / Codex / Claude Code

Point the MCP server command at node + absolute path to dist/index.js with arg serve.

The After Effects panel can write the Codex entry directly into C:\Users\YOU\.codex\config.toml. Restart Codex after installing it.

ChatGPT

ChatGPT uses a remotely reachable HTTPS MCP endpoint rather than launching this local Windows stdio command directly. Select ChatGPT (remote MCP) in the After Effects panel to create ~/.ae-mcp/chatgpt-setup.json and a setup guide. Set AE_MCP_CHATGPT_MCP_URL before setup if you already have an authenticated HTTPS relay. Never expose the local bridge without authentication.

CLI

Typical agent workflow

1. ae_health
2. ae_create_composition / ae_get_composition
3. ae_create_layer → ae_set_keyframes / ae_set_expression / ae_apply_effect
4. ae_capture_frame — look at the still
5. Iterate

Tool catalog (v0.1)

Architecture

MCP client → ae-mcp (stdio) → in-process broker → ~/.ae-mcp/instances/<id>/
                                                     command.json → result.json
                                              ← AE Startup bootstrap.jsx

Development

npm run build
npm test
npm run typecheck

Before publishing changes, run the full local release check described in the installation guide.

License

MIT