Which Indian language should I support first?

Hindi has the widest reach nationally; after that, prioritize by your actual user base — Tamil, Telugu, Marathi, and Bengali are common next choices.

Should tool names be translated?

No. Keep tool names in English for consistency across clients and integrations; translate descriptions and error messages instead.

How do I handle mixed-language input?

Use a language detector that tolerates code-switching, and fall back to English processing when detection confidence is low rather than guessing.

internationalizationTechArticle

MCP Multi-Language Support for Indian Languages

Detect the user's language, translate at the edges of a tool call, and localize error messages for Hindi, Tamil, Telugu, and other Indian languages.

Quick Answer / TL;DR

Detect the caller's language, translate the query to English at the boundary of a tool call so business logic stays language-agnostic, run the tool, then translate the result back — and keep tool names themselves in English for consistency across clients.

Key Takeaways

Translate at the edges of a tool call; keep the business logic itself language-agnostic.
Keep tool names in English always; localize descriptions and error messages only.
Normalize Unicode input to NFC form before processing to avoid subtle matching bugs.

Detect, translate at the edge, translate back

Detect the language of the incoming query, translate it to English before it reaches tool logic, run the tool as normal, then translate the result back to the caller's language. This keeps every tool implementation language-agnostic instead of branching internally per language.

typescript

server.setRequestHandler(CallToolRequestSchema, async (request) => {
  const { query, language } = request.params.arguments as { query: string; language?: string };
  const detected = language ?? detectLanguage(query);

  const englishQuery = detected !== "english" ? await translateText(query, "en") : query;
  const result = await executeTool({ ...request, query: englishQuery });

  if (detected !== "english") {
    return { content: [{ type: "text", text: await translateText(JSON.stringify(result), detected) }] };
  }
  return result;
});

Localized error messages

Error messages are worth hand-translating rather than machine-translating on the fly, since they are a small, fixed set and users read them under frustration where clarity matters most.

text

const ERROR_MESSAGES: Record<string, Record<string, string>> = {
  validation_error: {
    english: "Invalid input provided",
    hindi: "अमान्य इनपुट प्रदान किया गया",
    tamil: "தவறான உள்ளீடு வழங்கப்பட்டது",
  },
};