Automating AI Dubbing Pipelines With Ollang's API, Webhooks, SDK, and MCP Server

Most localization teams still move dubbing work by hand: someone downloads a mezzanine file, uploads it to a vendor portal, fills in a language list, waits, checks a dashboard, downloads deliverables, and re-uploads them to the content system. That works for ten titles a quarter. It does not work for episodic catalogs, weekly marketing drops, or accessibility mandates with fixed deadlines. The bottleneck is rarely the dubbing itself; it is the glue work between your content systems and the dubbing provider.

Ollang exposes its dubbing platform through an AI dubbing API, a REST API covering uploads, project and order creation, revisions, human review, and asset retrieval, plus webhooks, a TypeScript/Node.js SDK, and an MCP server. This guide walks through how to connect each stage of the dubbing lifecycle to those interfaces. One caveat up front: Ollang documents the API surface, not turnkey connectors. Named integrations with specific CMS, DAM, MAM, or OTT products are not verified, so plan on your team (or a partner) writing the integration code against the documented endpoints.

Map the Dubbing Lifecycle to System Events

Before writing any code, map your internal content events to the dubbing lifecycle Ollang's documentation describes:

1. Source ready, a master file, and optionally subtitles, an M&E track, glossaries, and voice instructions, exist in your storage.
2. Order placed, an AI dubbing order is created with target-language configuration.
3. Processing, transcription, translation, voice synthesis, optional lip sync, and audio mixing run inside Ollang's workflow.
4. Review, the order either stays AI-only but editable, or routes through assigned linguists and editors with review gates.
5. Delivery, mixed video, dubbed audio, vocals-only tracks, M&E, subtitles, and production files become retrievable.

Each of these stages has a corresponding API operation or webhook event. The integration pattern that follows is straightforward: your systems push at stages one and two, Ollang pushes back at stages four and five, and humans intervene only where your workflow configuration says they should.

The design decision that matters most is where review happens. Ollang supports AI-only workflows that remain editable and rerunnable, and AI-plus-human workflows with assignment to linguists, editors, Ollang-managed reviewers, or external LSPs. Decide per content type, a training video and a flagship series should not share the same gate configuration, and encode that decision in your order-creation logic rather than leaving it to per-project judgment.

Upload Media and Create Projects Programmatically

The upload endpoint supports direct media uploads of video files up to 30 GB, which covers most broadcast-grade mezzanine files without requiring you to transcode a proxy first. Documented input formats include MP4, MOV, AVI, WMV, FLV, and MKV for video, and WAV, MP3, AAC, M4A, FLAC, and OGG for audio. Script-only TTS workflows can start from a .txt file with no video at all.

A project is more than a single file. Ollang's asset model lets you attach:

• Subtitle references (SRT or VTT) that serve as timing and segmentation anchors for the dub.
• M&E tracks, upload a clean music-and-effects stem if you have one, or let the platform extract one from the source.
• Supporting materials, glossaries, brand guidelines, voice instructions, character lists, reference translations, and accessibility notes.

For libraries, the structured folder upload is the feature to build around: it creates multiple projects and associates source video, audio, subtitles, M&E, guidelines, and other assets automatically. Ollang documents this specifically for episodic content and large archives. If your MAM can export a folder structure matching Ollang's expected layout, an entire season becomes one bulk operation instead of dozens of manual uploads.

Authentication for the REST API uses API keys, so your integration service holds a key per environment and calls endpoints directly.

Create AI Dubbing Orders and Language Configurations

Once a project holds its assets, order creation is a single API call. The API exposes aiDubbing as an order type, with target-language configuration and optional rush flags. The same API also supports studio dubbing orders, so a pipeline can route premium titles to human performance and volume content to AI synthesis through the same interface.

Language configuration deserves attention. Ollang's workflow architecture lets speech-to-text, translation, and TTS providers be configured independently, by language pair, order type, organization-wide workflow, or folder-level workflow. In practice this means your integration does not hard-code model choices; it references a workflow, and a localization manager can retune provider routing for a difficult language pair without touching the pipeline code. One published caveat: Ollang's supported-languages endpoint documents translation and transcription coverage, not universal voice-synthesis availability, so validate dubbing availability for each target language you plan to automate before committing SLAs downstream.

Orders can also request optional capabilities documented in the AI dubbing workflow, such as lip sync or audio description, where those fit the content.

Trigger Review, Revision, and QC Operations

This is where an API integration earns its keep, because review is where manual pipelines stall. Ollang's documented review and editing workflow supports:

• Assigning orders to linguists, editors, Ollang-managed reviewers, or external LSPs, with restricted reviewer visibility.
• Segment-based dialogue editing: translation changes, pacing adjustments, and speaker management.
• Segment-level resynthesis after edits, so a corrected line regenerates without redoing the whole order.
• Rerunning workflows after revisions, with review gates controlling final delivery.

Platform-level QA capabilities include configurable review gates, human QC annotations, QC thresholds with automatic escalation to linguists, glossary and translation-memory enforcement, and analytics on QC-score progression and human-edit percentage. One scope limitation worth encoding into your expectations: Ollang's documented built-in AI QC scoring applies to subtitle translation orders. Do not assume equivalent automated scoring exists for synthesized voice quality, mixing, or lip-sync accuracy, for those, your gate is human review.

For the integration, the pattern is: create the order with the appropriate review configuration, then let assignment and escalation run inside Ollang while your system waits for events rather than polling a dashboard.

Use Webhooks to Track Status and Delivery

Webhooks are what turn this from an API wrapper into a pipeline. Ollang provides webhooks for workflow events, which lets your integration react instead of poll:

• Update the title's localization status in your CMS or project tracker when processing stages complete.
• Notify a Slack channel or ticketing system when an order hits a review gate that needs a decision.
• Trigger deliverable retrieval automatically when an order reaches final delivery.

On delivery, the retrieval endpoints give you access to the documented deliverable set: the mixed master video, dubbed audio, vocals-only dubbed audio, created or extracted M&E, source-vocals-only audio, embedded-subtitle video, and production files including dubbing scripts and dubbing SRTs. A delivery-event handler that fetches the right subset per content type, mixed master to the OTT ingest path, M&E and vocals to the audio archive, dubbing SRT to the subtitle repository, closes the loop without anyone downloading a file by hand.

Note that exact containers, codecs, and channel configurations for every AI dubbing output are not consistently published, so verify deliverable specs against your downstream ingest requirements during evaluation.

Choose Between REST, SDK, MCP, and Agent Skills

Ollang offers four integration surfaces. They are not competitors; pick per use case.

REST API. The full surface: uploads, orders, revisions, review, retrieval, and webhook configuration, authenticated by API key. Choose this for backend services in any language, or when you need every endpoint.

TypeScript/Node.js SDK. If your integration layer runs on Node.js, the SDK gives you typed calls against the same API. This matters for maintainability: typed order-creation and asset models catch integration errors at compile time rather than in production.

MCP server. Ollang provides an MCP server using OAuth 2.0 with PKCE. This is the path for AI-agent tooling, connecting an assistant or agent framework to dubbing operations with delegated, user-scoped authorization rather than a shared API key. The PKCE flow matters here because agent clients often cannot safely hold a static secret.

Agent Skills. Ollang also documents file-based agent Skills, a lighter-weight way to expose dubbing operations to agent environments that consume skill definitions.

For a production content pipeline, the typical shape is: REST or SDK for the automated backbone, webhooks for state, and MCP for interactive or agent-driven operations layered on top.

How to Evaluate and Get Started

Run a scoped pilot before committing pipeline code. A reasonable sequence:

1. Pick one content type and two target languages, and confirm dubbing availability for those languages with Ollang directly, since the published language list covers translation and transcription rather than guaranteed voice synthesis.
2. Test the upload path with a real mezzanine file, including subtitle reference and M&E if you have them, and confirm the 30 GB direct upload fits your masters.
3. Create an aiDubbing order via the API, exercise a revision cycle with segment-level resynthesis, and verify webhook events arrive for each stage you care about.
4. Inspect deliverables against your ingest specs, since output containers and codecs should be confirmed rather than assumed.
5. Ask Ollang directly about anything not in public documentation, turnaround expectations for your content profile, voice-cloning consent and enrollment processes, and any security requirements beyond the documented roles and SOC 2 claim.

The endpoints, SDK, webhooks, and MCP server are documented and available now. The integration work is yours to build, but it is glue code against a defined API, not a portal-scraping workaround, and that is the difference between dubbing as a per-title task and dubbing as a pipeline stage.