Skip to content

architecture.md

Latest commit

 

History

History
251 lines (207 loc) · 16.4 KB

architecture.md

File metadata and controls

251 lines (207 loc) · 16.4 KB

PremierPro MCP Server — Architecture

System Overview

The PremierPro MCP Server is a multi-language system that enables AI-driven end-to-end video editing in Adobe Premiere Pro. It uses four languages — Go, Rust, Python, and TypeScript — each chosen for their strengths in a specific layer of the stack.

Architecture Diagram

PremierPro MCP Server — System Architecture

CLI / MCP Client Claude Code · MCP Plugin · Custom CLI stdio / JSON-RPC Go — MCP Server & Orchestrator MCP Protocol Handler Task Orchestration Service Mesh Health & Recovery Concurrency (goroutines) Retry & Circuit Breaker Graceful Shutdown Logging & Metrics gRPC gRPC gRPC / HTTP Rust — Media Engine Media Probe & Metadata Extract Asset Indexing & Fingerprinting Waveform & Silence Detection Thumbnail Generation FFmpeg bindings · Zero-copy I/O Python — Intelligence Script Parsing & NLP Edit Decision List (EDL) Gen Shot ↔ Asset Matching (AI) Pacing & Timing Analysis LLM · Embeddings · Scene Detection TypeScript — PP Bridge ExtendScript API Layer CEP Panel (Primary Bridge) Timeline & Clip Operations Import / Export / Render CEP (primary) · Standalone Node (fallback) CEP Panel Node Standalone Adobe Premiere Pro Scripting DOM · Timeline · Media Browser · Render Shared Data Layer Asset Index (DB) EDL Cache Proto Definitions Config & Secrets Example Flow: "Edit video from script" 1. CLI Request "Edit with script.pdf" 2. Go Orchestrate Fan-out to services 3a. Rust: Scan assets 3b. Python: Parse script 4. TS: Execute EDL Timeline assembly 5. Premiere Pro Render final output Go Rust Python TypeScript Premiere Pro ──── gRPC - - - Flow

Language Responsibilities

Language Layer Responsibilities
Go Orchestrator & MCP Server MCP protocol (JSON-RPC/stdio), task orchestration via goroutines, service mesh, health checks, retry/circuit-breaker, graceful shutdown, logging & metrics
Rust Media Processing Engine Media probing & metadata extraction, asset indexing & fingerprinting, waveform & silence detection, thumbnail generation, high-perf file I/O via FFmpeg bindings
Python Intelligence Layer Script parsing & NLP, Edit Decision List generation, shot-to-asset matching via AI embeddings, pacing & timing analysis, scene detection
TypeScript Premiere Pro Bridge ExtendScript API integration, CEP Panel (primary bridge), standalone Node.js fallback, timeline & clip operations, import/export/render control

Inter-Service Communication

All services communicate via gRPC with shared protobuf definitions:

Route Protocol Payload
CLI → Go stdio / JSON-RPC 2.0 MCP tool calls
Go → Rust gRPC (protobuf) Media scan requests, asset queries
Go → Python gRPC (protobuf) Script text, EDL generation requests
Go → TypeScript gRPC / HTTP Premiere Pro commands (EDL execution)
TypeScript → Premiere Pro CEP / ExtendScript Native Adobe scripting DOM calls

Bridge Fallback Strategy

The TypeScript bridge to Premiere Pro supports two modes:

  1. CEP Panel (Primary) — Runs inside Premiere Pro as an extension panel. Direct DOM access, lowest latency. Communicates with the Go orchestrator over a local WebSocket/HTTP server.

  2. Standalone Node.js (Fallback) — Runs as an external process. Sends commands to Premiere Pro via the ExtendScript Toolkit CLI (osascript / COM on Windows). Higher latency but works without a panel installed.

The Go orchestrator auto-detects which bridge is available and falls back gracefully.

Build System

A unified build using just (justfile) or make that:

  • Compiles the Rust engine (cargo build)
  • Builds the Go orchestrator (go build)
  • Installs Python dependencies (pip install / poetry)
  • Bundles the TypeScript bridge (npm run build)
  • Generates protobuf stubs for all four languages
  • Runs all test suites
  • Packages the CEP panel for Premiere Pro installation