Read this guide in العربية or Français.
Claude Code can write video components, arrange Remotion scenes, and run FFmpeg, but it cannot produce a convincing Algerian voice by itself. Natiq now provides a remote MCP server and an open-source Claude Code plugin that add Algerian Darija speech generation directly to that workflow.
Once connected, Claude can discover Algerian voices, generate one audio clip per scene, wait for asynchronous jobs, and download the finished WAV files into your project. More importantly, it can build the video around the measured audio instead of guessing how long a written script will take to speak.
What the Natiq integration gives Claude
The plugin connects Claude to tools that can:
- list available Algerian voices and delivery purposes;
- check the account balance before a large job;
- submit speech generation for each scene;
- poll a pending job without spending tokens again;
- recover recent jobs after an interrupted response; and
- return a temporary download URL and the real
duration_seconds.
The MCP server is hosted at https://natiq.studio/mcp. It uses the same API keys, scopes, rate limits, token billing, and internal generation services as Natiq's public REST API.
The easiest way to connect
Open the Claude Code connection guide and select Copy setup prompt. Paste that prompt into Claude Code. Claude will install the public plugin at user scope and verify that the bundled natiq MCP server is present.
For manual installation, run this inside Claude Code:
/plugin install natiq-voiceover --marketplace Natiq-studio/natiq-voiceover
Create a write-scoped key on the API keys page, then enter it once in the plugin's masked Natiq API key option. Claude stores sensitive plugin configuration in the secure credential store provided by your operating system.
A scene-by-scene video workflow
After connecting, a useful request looks like this:
Make a three-scene Remotion product video in Algerian Darija. Check my balance, use one consistent voice, save audio in public/audio, and time every scene from the actual clip duration. Reuse clips whose text and settings have not changed.
Claude can then follow a reliable sequence:
- Check the balance and available voices.
- Split the message into a short narration script for each scene.
- Submit each scene once and record its
speech_...job ID. - Poll the same job until it succeeds instead of creating a second paid job.
- Download the WAV clips into the video repository.
- Set scene lengths from the duration returned with each successful result.
Timing Remotion scenes from real audio
For a 30 fps composition, a scene can use this calculation:
const fps = 30;
const durationInFrames = Math.ceil(duration_seconds * fps) + 6;
The extra six frames provide a short visual pause. Use the measured value from Natiq after the generation job succeeds. Word count is not a reliable substitute because speech rate, punctuation, and delivery style all change the final length.
Downloading audio safely
A remote MCP server cannot write directly to a user's computer. Natiq therefore returns a temporary signed URL that Claude can download into the project:
mkdir -p public/audio
curl --fail --location --output public/audio/scene-01.wav "RETURNED_AUDIO_URL"
The signed URL works without attaching the API key. Download the file promptly and keep temporary URLs out of Git; the rendered project should reference the local audio file.
Caption timing is an estimate
Natiq returns total clip duration, not word-level timestamps. Claude can create scene-level captions or distribute words approximately across the clip, but that timing must be labelled as estimated. Accurate karaoke-style highlighting requires a separate forced-alignment or transcription step after generation.
Practices that save time and tokens
- Review the complete scene script before generation.
- Keep one voice and delivery style across a video unless the story needs multiple speakers.
- Store job IDs, text, voice, and purpose in the scene manifest.
- Reuse a downloaded clip when its narration and settings have not changed.
- Check the balance before generating a large batch.
- Poll pending jobs instead of resubmitting unchanged text.
For more implementation detail, read the Algerian Darija TTS API guide. To start from the simplest workflow, open the Claude Code connection page, copy the setup prompt, and let Claude install Natiq for you.