MCP for Localization: How AI Agents Can Manage Your Translations

When you're building a feature and adding a new UI string, the traditional workflow looks like this: write the code, add a placeholder string, commit, open your TMS dashboard, manually add the new key, write the English source string, request translations, wait, come back later to check, publish. The disconnect between writing code and managing translations is a constant source of friction.

The Model Context Protocol (MCP) offers a different model. Instead of switching between your coding environment and a separate dashboard, your AI assistant — the same one helping you write code — can manage your translations directly, in the same conversation.

This is what Better i18n's MCP server enables.

What Is MCP?

The Model Context Protocol is an open standard developed by Anthropic that lets AI models interact with external tools and data sources in a structured way. An MCP server exposes a set of "tools" that an AI assistant can call — similar to how a web API exposes endpoints.

When you configure Claude (or another MCP-compatible AI) to use the Better i18n MCP server, Claude gains the ability to:

• List your translation projects and keys
• Read current translations
• Add new translation keys
• Propose translations for missing languages
• Publish approved translations
• Check sync status with GitHub

All of this happens in natural language, within your development conversation.

Setting Up Better i18n MCP

Installation takes about 2 minutes. Add the MCP server to your configuration:

// .mcp.json (in your project root or ~/.mcp.json globally)
{
  "mcpServers": {
    "better-i18n": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@better-i18n/mcp@latest"],
      "env": {
        "BETTER_I18N_API_KEY": "your-api-key"
      }
    }
  }
}

That's it. Restart Claude, and it now has access to your Better i18n workspace.

What You Can Do With It

Check what's missing

You: "What translation keys are missing German translations in the onboarding project?"

Claude: [checks better-i18n]

There are 12 keys missing German translations:
- onboarding.welcome.title
- onboarding.welcome.subtitle
- onboarding.step1.heading
...

Would you like me to propose German translations for these?

Add new keys while writing code

You: "I'm adding a new error state for when the API rate limit is exceeded.
      Add a translation key for 'You've reached your API rate limit.
      Upgrade your plan to continue.'"

Claude: Added key `errors.api.rate_limit_exceeded`. Also proposed translations
for your 6 active languages:
- de: "Sie haben Ihr API-Limit erreicht. Aktualisieren Sie Ihren Plan..."
- fr: "Vous avez atteint votre limite API. Mettez à jour votre abonnement..."
- es: "Ha alcanzado su límite de API. Actualice su plan para continuar."
- ja: "APIレート制限に達しました。続けるにはプランをアップグレードしてください。"

Should I publish these, or would you like to review them first?

Review and publish selectively

You: "German and French look good. Hold Japanese — we need a native reviewer.
      Publish the others."

Claude: Published de, fr, es translations for `errors.api.rate_limit_exceeded`.
Japanese is queued as draft pending review.

The Developer Workflow Shift

The fundamental shift MCP enables is colocation of code and localization work. When translation management is a separate dashboard you switch to, it creates cognitive overhead and a tendency to defer localization. Strings go untranslated. Keys accumulate without translations.

Before MCP:

1. Write feature code
2. Use placeholder string or hardcode English
3. Commit
4. Open TMS dashboard (separate tab, separate context)
5. Add the key, write English source, request translation
6. Wait for turnaround, come back, publish

With MCP:

1. Write feature code
2. Tell Claude: "Add translation key for this string with AI translations for all active languages"
3. Review in the same conversation
4. Publish or queue for review
5. Continue coding

The localization step goes from "deferred task" to "part of the coding workflow."

Use Cases Beyond Development

MCP isn't just for engineers. The same tools can be used by:

Product managers checking translation coverage before a launch:

"Give me a translation coverage report for the billing module across all 8 languages"

Content teams managing marketing copy:

"Which marketing page keys are missing Spanish translations?"
"Propose Spanish translations for all missing homepage keys"

Customer success handling urgent fixes:

"There's a typo in the French welcome email subject line.
 Fix it: 'Bienvenue chez Better i18n' should be 'Bienvenue sur Better i18n'"

What the MCP Server Exposes

The @better-i18n/mcp package exposes the following tools:

The MCP server is open source (packages/mcp in the OSS repo), so you can inspect exactly what it does, extend it, or contribute improvements.

Privacy and Security

API key scoping: Your MCP API key can be scoped to read-only, draft-only, or full access. For AI tools, we recommend draft-only in production environments — the AI can propose translations, but a human must publish.

Audit log: Every action taken by the MCP server is logged in your Better i18n workspace with the API key that performed it.

Review workflow: AI-proposed translations are created as drafts by default. They go through the same review workflow as human-proposed translations before being published.

The Bigger Picture

MCP is an early signal of a broader shift in how software tooling works. The traditional model — humans switch between tools, copy-paste between contexts, manage state in their heads — is being replaced by AI-orchestrated workflows where the AI holds context and coordinates between tools.

Localization is a perfect example. It's high-volume, structured, requires coordination between code and content, and benefits enormously from AI assistance. The teams that integrate AI deeply into their localization workflow will move significantly faster in international markets.

Set up Better i18n MCP in 2 minutes →