Crowdin Machine Translation: Add Custom MT Engine (2026 Guide)

Crowdin supports multiple machine translation engines out of the box: Google Translate, Microsoft Translator, DeepL, and several others. But the built-in options aren't your only choices. Crowdin's custom MT engine feature lets you plug in any translation API that follows their specification.

This is useful for two reasons. First, you might want a cheaper engine than Google or DeepL. Second, you might want a specialized engine that handles your content type better than general-purpose MT. This guide walks through the setup process and shows how to use Langbly (or any Google Translate v2 compatible API) as a Crowdin MT engine.

How Crowdin uses machine translation

Before diving into setup, it helps to understand how Crowdin actually uses MT engines:

Pre-translation

The most common use case. When new source strings are added to your project (via Git sync, file upload, or API), Crowdin can automatically translate them using your configured MT engine. Translators then see the machine-translated version as a suggestion they can accept, edit, or replace.

Pre-translation saves significant time. Instead of translating from scratch, translators start with a first draft that's often 80-95% correct (depending on the engine and content type). Their job becomes reviewing and polishing rather than creating from nothing.

Translation suggestions

While a translator works on a string, Crowdin shows MT suggestions in the sidebar alongside translation memory matches and glossary terms. The translator picks the best starting point.

Cost per suggestion

Crowdin charges nothing for the MT feature itself, but you pay for the translation API calls. Pre-translating 10,000 strings at an average of 50 characters per string = 500,000 characters. At Google Translate's $20/M rate, that's $10. At Langbly's $1.99-$3.80/M rate, it's $1-$2.

Multiply by 10 languages and you see why MT engine pricing matters. Pre-translating a 10,000-string project into 10 languages costs $100 with Google Translate or $10-$20 with Langbly.

Setting up a custom MT engine in Crowdin

Step 1: Navigate to MT engines

In your Crowdin project, go to Settings > Machine Translation. (In Crowdin Enterprise, this is under Resources > Machine Translation Engines at the organization level.)

Step 2: Add a custom engine

Click "Add MT Engine" and select the engine type. For built-in engines (Google, DeepL, Microsoft), you just enter your API key. For a custom engine, select "Custom MT" or the specific format your engine supports.

Langbly works as a Google Translate v2 compatible engine. In Crowdin, you can configure it as a Google Translate engine with a custom endpoint URL. Here's how:

Select "Google Translate" as the engine type
Enter your Langbly API key (get one from your Langbly dashboard). See the quickstart guide for setup details.
Set the API URL to https://api.langbly.com
Save the configuration

Because Langbly uses the same v2 request/response format as Google Translate, Crowdin treats it identically. The API receives the same requests and returns the same response structure.

Step 3: Test the connection

After saving, Crowdin shows a "Test" button. Click it to verify the engine responds correctly. If the test fails, check:

API key is correct and active
API URL is correct (include https://, no trailing slash)
Your Langbly account has available character credits

Step 4: Set the engine as default

If you have multiple MT engines configured, set your preferred one as the default. The default engine is used for pre-translation and suggestions unless overridden per language pair.

You can also set different engines for different language pairs. This is useful if one engine is stronger for European languages and another for Asian languages.

Pre-translation workflow

Once your MT engine is configured, set up automatic pre-translation:

Manual pre-translation

Go to your project, select the files or folders you want to pre-translate, and click "Pre-translate." Choose:

Translation engine: Your configured MT engine
Languages: Which target languages to pre-translate
Overwrite: Whether to overwrite existing translations (usually "no" to preserve human work)
Approval: Whether to auto-approve pre-translated strings (usually "no" for MT output)

Automatic pre-translation

For continuous localization, enable automatic pre-translation on new strings. In project settings:

Go to Settings > Pre-translation
Enable "Pre-translate new strings"
Select your MT engine
Choose which languages to include
Set the approval status for pre-translated strings (recommend: "Unapproved" so translators review them)

Now, when a developer pushes new strings via the Git integration, Crowdin detects them, sends them to your MT engine, and populates the translations automatically. By the time a translator opens Crowdin, first drafts are already waiting.

Optimizing quality

Use translation memory first, MT second

Configure Crowdin to check translation memory before falling back to MT. Exact TM matches are pre-approved human translations. They're guaranteed consistent. MT output needs review.

In pre-translation settings, set the priority: TM exact match > TM fuzzy match (above 90%) > MT. This way, strings that have been translated before reuse the approved version, and only genuinely new strings get machine-translated.

Set up glossaries

Upload your terminology glossary to Crowdin. When translators review MT suggestions, the glossary helps them verify that product-specific terms are translated correctly. Some MT engines handle terminology better than others.

Context-aware translation engines like Langbly tend to handle domain-specific vocabulary better than traditional NMT because they understand the surrounding context. But a glossary is still valuable as a reference for your reviewers.

Monitor and improve

Crowdin tracks how often MT suggestions are accepted, modified, or rejected. Check these statistics regularly:

High acceptance rate (80%+): Your MT engine is performing well for this language pair
Moderate acceptance (50-80%): Working but with room for improvement. Check which error types cause edits.
Low acceptance (below 50%): The MT engine struggles with this content or language. Consider a different engine for this pair.

Cost optimization with Crowdin + Langbly

The biggest cost in a Crowdin translation workflow isn't the TMS subscription. It's the MT engine usage and translator time. Reducing either one saves money.

Langbly reduces MT costs directly: $1.99-$3.80 per million characters versus $20-$25 for Google or DeepL. Check our pricing page for current plans. For a project with 20,000 strings across 10 languages, pre-translation costs:

Engine	Cost per pre-translation run	Annual cost (monthly updates)
Google Translate	$20	$240
DeepL Pro	$25 + $5.49/mo	$366
Langbly (Growth)	~$2.76	$33

But the indirect savings matter more. If Langbly's context-aware translations are accepted by translators 85% of the time instead of 70% for a basic NMT engine, translators spend less time editing. At $0.10 per word for translation/review, a 15% improvement in MT acceptance rate across 200,000 words saves $3,000 per language per year.

For a full cost analysis, see the cheapest translation API comparison and the Google Translate pricing guide.

Crowdin Enterprise vs. Crowdin (standard)

The MT integration works similarly in both versions, with a few differences:

Standard Crowdin: MT engines are configured per project. Custom engines are available on Team plan and above.
Crowdin Enterprise: MT engines are configured at the organization level and shared across projects. Custom engines are available on all Enterprise plans.

If you're using Crowdin Standard on the Free plan (open source), custom MT engines may not be available. Check your plan's feature list. The built-in Google Translate option works on all plans, which means the "configure as Google Translate with custom URL" approach for Langbly works on any plan that supports Google Translate.

Other TMS platforms

This guide focuses on Crowdin, but the concept applies to other translation management systems too. Lokalise, Phrase, and Transifex all support custom MT engines. The setup differs per platform, but the workflow is the same: configure the engine, set up pre-translation, have translators review the output.

For the broader picture of how TMS fits into your localization workflow, read the software localization guide.

Crowdin Machine Translation: How to Add a Custom Translation Engine