Algerian Darija Voiceovers from Claude Code: Connecting the Natiq MCP Server
If you make videos with Claude Code (Remotion compositions, motion graphics, ffmpeg pipelines), the voiceover is usually the step that pulls you out of the terminal. You write the script, paste it into a TTS website, dow
If you make videos with Claude Code (Remotion compositions, motion graphics, ffmpeg pipelines), the voiceover is usually the step that pulls you out of the terminal. You write the script, paste it into a TTS website, download the files, rename them, and then guess how long each scene should run.
For Algerian Darija that step is worse, because most speech tools read Darija as Modern Standard Arabic or as Moroccan. Natiq Studio is a text-to-speech service built for Algerian Darija, and it now exposes a remote MCP server, so Claude Code can write the narration, generate the audio, download it into your project and time each scene to the real clip length.
I built Natiq, so read this as a maker's write-up, not a neutral review.
What the server does
The endpoint is https://natiq.studio/mcp. It speaks Streamable HTTP and authenticates with a normal Natiq API key. It gives Claude six tools:
| Tool | What it's for |
|---|---|
list_voices |
Current voice IDs, with preview audio |
list_purposes |
Delivery styles (ad, narration, and so on) |
check_balance |
Remaining tokens and pricing |
generate_speech |
Submit one scene; spends tokens once |
get_speech |
Poll a job and get a fresh audio URL |
list_speech |
Find recent jobs after an interrupted response |
There is no batch endpoint on purpose. One scene per call keeps spending visible, and if a call fails halfway you lose one scene, not the whole video.
Setup in three steps
The full walkthrough, with copy buttons, is on the Connect Natiq to Claude page.
1. Create an API key. Sign up at natiq.studio and create a write-scoped key. A read-only key can list voices and check your balance, but it cannot generate audio.
2. Install the plugin. In Claude Code:
/plugin install natiq-voiceover --marketplace Natiq-studio/natiq-voiceover
Confirm the marketplace and choose User so it works in every project. The plugin is MIT licensed and lives at github.com/Natiq-studio/natiq-voiceover. It adds the MCP connection plus a voiceover skill that tells Claude how to split a script into scenes and reuse unchanged clips.
3. Paste the key when prompted. Claude Code masks it and stores it in your system's credential store. Run /mcp afterwards; Natiq should show as connected.
If you'd rather skip the plugin, one command does the connection alone:
claude mcp add --transport http natiq https://natiq.studio/mcp --header "Authorization: Bearer natiq_live_..."
Use one or the other. With both, Claude sees every Natiq tool twice.
Asking for a video
After setup you describe the video in plain words. For example:
/natiq-voiceover:voiceover Make a two-scene product video in Algerian Darija.
Check my balance, use one voice, save the audio in public/audio, and time
each scene to the actual clip duration. Reuse any unchanged clips.
Claude writes each scene in Darija, calls generate_speech once per scene, polls get_speech until the clip is ready, downloads the WAV into public/audio, and reads duration_seconds from the response to set the scene length in your composition.
That last part is the reason I wanted this inside Claude Code. Darija sentences mixing in French words run longer or shorter than you'd guess from the character count, and a scene timed by estimate either cuts the speaker off or leaves dead air. Timing from the measured duration fixes that.
Things that behave differently from what you might expect
-
Generation is asynchronous.
generate_speechreturns a job ID, often stillpending. That is not an error. The skill tells Claude to wait at least two seconds between polls and respectretry_after. -
No idempotency key. If a network error hides whether a submit went through, Claude should check
list_speechbefore trying again, or you pay twice for the same scene. -
Audio URLs expire. They are signed links. Download them right away and don't commit them to a repository.
get_speechhands out a fresh one when you need it. - No word timestamps. Scene timing is exact; caption timing inside a scene is an estimate unless you run a separate aligner.
- It spends real tokens. Each clip is billed from your account balance at the same rates as the Natiq API and the web studio.
Who it's for
Mostly two groups: Algerian agencies and e-commerce sellers who already script product videos in Remotion or ffmpeg and want the Darija voiceover in the same loop, and developers building tools for Algerian users who want to try Darija speech without writing an API client first. If you only need a single clip now and then, the web studio at natiq.studio is quicker.
Links
- Connection guide: natiq.studio/claude
- API and MCP reference: natiq.studio/api/docs
- Plugin source: github.com/Natiq-studio/natiq-voiceover
- Natiq Studio: natiq.studio
Disclosure: I'm the developer of Natiq Studio. I drafted this article with AI assistance and checked the commands against the live service.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.