Premiere Pro MCP

Control Adobe Premiere Pro with natural language through Claude (or any MCP-compatible LLM).

What it does

You: "Apply a cinematic look to all clips on V1, ducking the music
      when there's voice, and export to YouTube 1080p."

Claude: ✓ Applied Lumetri (-15K temperature, +10 contrast) to 4 clips
        ✓ Added voiceover ducking on A2 (-12dB during A1 voice)
        ✓ Queued H.264 1080p export to ~/Videos/final.mp4

437 tools exposed via MCP to drive Premiere Pro 2026 from any LLM agent. Tested in runtime against PP 2026.2.2 on Spanish-localized Windows install.

Highlights

• Locale-resilient: works on Premiere Pro in Spanish, German, French, etc. (most competitors hardcode English component names and silently fail)
• Honest errors: when a Premiere host API genuinely doesn't expose something (PP 2026 has known limitations), you get a clear error with the recommended workaround — not a silent no-op
• Battle-tested: 137 tools validated in live runtime against PP 2026 with detailed bug fixes (see CHANGELOG.md)
• AI-first: built for LLM-driven workflows. Pair it with Whisper for speech-to-text, ComfyUI for thumbnails, FFmpeg for analysis — chain MCPs

Quick start

1. Install the CEP plugin

Copy cep-plugin/ to your Adobe CEP extensions folder:

• Windows: %APPDATA%\Adobe\CEP\extensions\MCPBridgeCEP\
• macOS: ~/Library/Application Support/Adobe/CEP/extensions/MCPBridgeCEP/

Enable CEP debugging:

# Windows
reg add HKCU\Software\Adobe\CSXS.11 /v PlayerDebugMode /t REG_SZ /d 1 /f

2. Build the MCP server

cd premiere-pro-mcp
npm install
npm run build

3. Register with Claude Desktop

Add to %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "premiere-pro": {
      "command": "node",
      "args": ["C:/path/to/premiere-pro-mcp/dist/index.js"]
    }
  }
}

4. Open Premiere Pro

• Launch Premiere Pro 2026
• Window > Extensions > MCPBridgeCEP (the bridge panel)
• Click "Start Bridge"

5. Talk to Claude

"Show me the active sequence structure."
"Add a 1-second fade-in to the audio clip on A2."
"Apply Cross Dissolve at every cut on V1, 0.5 seconds each."

What works (validated runtime, PP 2026 es-ES)

Project & Inspection

• Full project / sequence / clip overviews
• Bins, markers, metadata (XMP), color labels
• Effect properties with all values, keyframes, source media details
• Find/search project items, list sequences

Timeline editing

• Add/remove/move/duplicate clips, split, trim
• Set in/out points, playhead, work area
• Selection by name/range/color, invert, deselect
• Enable/disable clips, link/unlink

Transform & Motion

• Position, scale, rotation, opacity (locale-resilient)
• Anchor point, blend mode
• Position/scale keyframes via add_keyframe

Color

• Lumetri full parameter control: exposure, contrast, highlights/shadows, whites/blacks, temperature/tint, saturation, vibrance, sharpen, vignette, faded film
• LUTs apply
• Copy/paste color grade across clips
• ⚠️ See limitations: applying Lumetri for the first time requires manual UI

Audio

• Per-clip volume, fade in/out (locale-resilient)
• Volume/panning keyframes
• Audio FX pipeline: Compressor, EQ, De-Esser, De-Hum, De-Reverb, Limiter, Noise Reduction
• Voiceover ducking
• Normalize, detect clipping

Effects & Transitions

• 60+ video effects (Gaussian/Directional Blur, Sharpen, Warp Stabilizer, Noise Reduction, Mirror, Spherize, Corner Pin, Lens Distortion, etc.)
• Transitions with locale-aware name lookup (Cross Dissolve works on any language install)
• Batch apply across cuts

Captions

• Import SRT files as caption tracks
• Pipeline integration: Whisper-transcribe → SRT → MCP creates captions

Export

• Single sequence + batch export via Adobe Media Encoder
• Match Source presets, YouTube/Vimeo/Twitter/Facebook presets
• Encode arbitrary files via AME
• AAF, OMF, FCPXML export

Multicam

• Create multicam sequences, switch angles, flatten

Markers

• Add/update/delete on sequence or clip
• Color labels, comments, in/out range
• Import/export CSV

Known limitations (PP 2026 host)

These are not bugs in this MCP — they're verified limitations of Premiere Pro 2026's ExtendScript/QE-DOM surface. Each tool returns a clear actionable error with the recommended workaround.

| Feature | Status | Workaround | |---------|--------|------------| | capture_frame programmatic | ❌ qeSeq.exportFramePNG reports success but never writes file | UI: File > Export > Media > Single Frame | | set_track_volume | ❌ Audio track DOM no longer exposes .components or .volume; QE has no setVolume | Use add_audio_keyframes per-clip, or Audio Track Mixer | | speed_change programmatic | ❌ qeClip.setSpeed signature changed in PP 2026 (rejects all known forms) | UI: Clip > Speed/Duration, or change duration with trim_clip | | Apply Lumetri for the first time | ❌ addVideoEffect reports success but doesn't add component | Apply Lumetri manually once; afterwards all 28 setters work perfectly | | LUFS integrated measurement | ❌ Not scriptable | Export and measure with ffmpeg-loudnorm | | Native PP Speech-to-Text | ❌ Adobe AI STT not exposed | Use Whisper + create_caption_track (works great) | | Adobe Libraries MOGRT | ❌ Adobe Libraries not scriptable | Use local .mogrt files via import_mogrt | | Dynamic Link with AE | ❌ import_ae_comps not implemented | Render from AE and import as video | | Async export completion | ⚠️ AME runs async, no callback | Poll get_render_queue_status |

Architecture

┌─────────────┐         ┌──────────────┐         ┌─────────────┐
│   Claude    │         │   Node.js    │         │  Premiere   │
│  (or LLM)   │◄───────►│  MCP server  │◄───────►│   Pro CEP   │
└─────────────┘  stdio  └──────────────┘  files  │   plugin    │
                                                  │   (panel)   │
                                                  └─────────────┘
                                                        │
                                                        ▼
                                                  ┌──────────────┐
                                                  │ ExtendScript │
                                                  │  DOM + QE    │
                                                  └──────────────┘

The bridge automatically prepends locale-resilient helpers to every script:

__findIntrinsic(clip, "motion" | "opacity" | "volume" | "lumetri" | "timeRemap")
// tries matchName ("AE.ADBE Motion") first — locale-INDEPENDENT —
// then localized displayName ("Movimiento", "Bewegung", "Mouvement", etc.)

__findProp(component, ["Position", "Posición", "Position", "Position"])
// also descends one level into nested groups (needed for Lumetri's
// grouped parameters like Exposure inside Basic Correction)

__readMaybe(obj, "isMuted", false)
// reads members that are property in some PP versions, method in others

__findTransition(name, "video")
// resolves transition names across locales:
// "Cross Dissolve" matches "Disolvencia cruzada" / "Fondu enchaîné"

Development

npm run dev      # Watch mode
npm run build    # Build dist/
npx tsc --noEmit # Type-check only

When adding a tool, prefer:

• __findIntrinsic / __findComponent over hardcoded displayName === "Motion"
• __readMaybe for any isSomething() / isSomething property
• Pass .ticks strings (not Time objects) to insertClip/setInPoint/etc.

Credits

Built on top of work from AdobePremiereProMCP with extensive fixes for PP 2026 compatibility, locale resilience, and runtime-validated behavior. See CHANGELOG.md for the full fix log.

License

MIT