MCP Server Setup

@shipi18n/mcp brings i18n quality assurance to any Model Context Protocol client — Claude Desktop, Cursor, or your own. The validation tools need no API key; translation needs your own provider key.

Configure your client

Add the server to your MCP client configuration:

{
  "mcpServers": {
    "shipi18n": {
      "command": "npx",
      "args": ["-y", "@shipi18n/mcp"]
    }
  }
}

Restart the client — most read their config only at launch — and the tools appear.

What needs a key, and what doesn't

Validation — no key

check_locales, check_glossary, diff_locales and check_placeholders compare your files and call no model at all. They work with the config above exactly as written — no env block, nothing to pay for.

review_locales is the meaning-level review, and it also needs no key: the server returns the translation pairs plus the review criteria, and your agent judges them using the model it is already running. Nothing is sent to us and nothing extra is billed.

Translation — your own key

translate_json and translate_file call an LLM, so they need your provider key passed through the client config:

{
  "mcpServers": {
    "shipi18n": {
      "command": "npx",
      "args": ["-y", "@shipi18n/mcp"],
      "env": { "ANTHROPIC_API_KEY": "your-key-here" }
    }
  }
}

Using it

Ask in plain language — the client picks the tool:

"Translate locales/en.json into Spanish and French"
"Check whether locales/de.json dropped any placeholders"
"What language codes do you support?"

Version note

Use 2.0.1 or newer. Version 2.0.0 exited silently when started through itsbin and has been deprecated on npm.