Translate Inside Claude, Cursor, and Claude Code with the Adara MCP Server

Today we're announcing adara-mcp-translate, an open-source Model Context Protocol server that drops Adara Translate directly into Claude Desktop, Cursor, Claude Code, and any other MCP-compatible AI agent. The package is live on npm (npmjs.com/package/adara-mcp-translate) — one config block, one API key, and your assistant gains four translation tools it can call from inside any conversation.

API or MCP — Which Should You Use?

Adara has had a public REST API for a while (see our translation API comparison). The MCP server is a different surface for a different operator:

• Use the REST API when you're writing code — a backend service, a CI/CD job, a Zapier-style automation, anything that runs without a human in the loop.
• Use the MCP server when an AI agent is the operator — drafting marketing copy in Claude, reviewing a localized document in Cursor, building a multilingual feature in Claude Code.

Both share the same backend, the same five engines (DeepL, Google, Microsoft, GPT, Gemini), and the same credit pool — so you can mix them freely with one API key.

Why MCP?

Model Context Protocol is the open standard Anthropic introduced for letting AI assistants discover and call external tools. Instead of copy-pasting a paragraph into a translator tab and pasting the result back, your agent calls the translator directly and keeps working. The same pattern applies to document translation, audio transcription, and credit lookups.

For localization teams, this means an editor reviewing a draft in Claude can ask "translate this section into French using DeepL" and get the result in the same chat — drawing on the team's pooled Adara credits, with one invoice and one audit trail.

What Your Agent Can Do

Once connected, the server exposes four tools to your assistant:

• translate_text — translate a string using DeepL, Google, Microsoft, GPT, or Gemini. Switch engines per request.
• translate_document — translate a local TXT, PDF, DOCX, XLSX, CSV, PPTX, or SRT file (up to 5 MB) and optionally write the result to disk. Defaults to DeepL.
• transcribe_audio — transcribe a local MP3, WAV, OGG, M4A, or WebM file (up to 25 MB) via Whisper.
• get_user_credits — read-only balance check. Free, doesn't consume credits.

Quick Setup

You'll need Node.js 18+ on the machine running your MCP client and an Adara API key (generate one at /dashboard/api-keys — new accounts get free credits on signup, no card required).

Claude Desktop and Cursor

Add this block to claude_desktop_config.json (or ~/.cursor/mcp.json for Cursor) and restart the client:

{
  "mcpServers": {
    "adara": {
      "command": "npx",
      "args": ["-y", "adara-mcp-translate"],
      "env": { "ADARA_API_KEY": "adara_live_xxxxxxxxxxxxxxxx" }
    }
  }
}

Claude Code

One command:

claude mcp add adara -e ADARA_API_KEY=adara_live_xxxxxxxxxxxxxxxx -- npx -y adara-mcp-translate

For OS-specific config paths and troubleshooting, see the full guide: Connect Adara to Claude Desktop, Cursor, or Claude Code.

Example Prompts

• "Translate 'The quick brown fox jumps over the lazy dog' into Spanish using DeepL."
• "Translate /Users/me/Documents/contract.docx into French and save it next to the original."
• "Transcribe /Users/me/Recordings/standup.m4a and summarize the action items."
• "Check my Adara credit balance before we start this batch."

Tip: use absolute paths in your prompts. The server does not expand ~, so ~/Documents/file.docx won't resolve — write /Users/yourname/Documents/file.docx instead.

Security Model

The server runs locally on your machine over stdio (the standard input/output pipe used by command-line tools) — no network port is opened on your machine, and your MCP client spawns its own subprocess. Your API key is never logged in full; only a 16-character prefix is printed at startup so operators can confirm which key is loaded. When you supply an outputPath for a document translation, the server refuses to overwrite existing files and rejects .. path traversal before any billed work runs.

Get Started

Generate a key at Dashboard → API Keys, paste the config block above into your MCP client, and ask your assistant to translate something. If you'd rather start from your code first, the translation API comparison covers pricing, code samples, and how Adara compares to running DeepL, Google, Microsoft, and GPT directly. For team-wide automation patterns that mix the two, see developer workflows.