Labsco
MCP SERVER

Lara Translate

by translated ↗

Translate with your own terminology enforced — glossaries, translation memories, tone and context — instead of hoping a general model gets the industry words right.

Document Conversion, PDF & Translation✓ Verified
Summary
Consistency across documents is the part a general model cannot give you.

A glossary makes a product name or a legal term land the same way every time, and a translation memory means a segment approved once comes back approved rather than re-invented — both applied on the translate call itself rather than in a post-edit. Set-up is the lightest part: the hosted endpoint authenticates through the browser with no key to place, and TMX and CSV imports let existing terminology come across instead of being retyped, though both return an import id you have to poll.

What it is

Translated's Lara Translate as an MCP server, hosted or self-run. Alongside translation it manages the two assets that make professional translation repeatable: translation memories of previously approved segments, and glossaries that pin specific terms.

What you get
  • translate takes source and target languages, surrounding context, a style of faithful, fluid or creative, short localization instructions, and the memories and glossaries to apply
  • detect_language handles a single string or up to 128 at once, returning the detected language, content type and confidence scores
  • list_languages returns the language codes the account supports
  • Translation memories created, renamed, listed and deleted, with individual source-target units added and removed — including the sentences before and after, which is what disambiguates a repeated segment
  • import_tmx loads an existing TMX file, returning an import id to poll with check_import_status and a progress field to watch
  • Glossaries created, renamed, listed, read and deleted, with entries added, replaced and removed in monodirectional or multidirectional form
  • import_glossary_csv loads terminology in bulk and polls the same way; export_glossary writes it back out as CSV, and get_glossary_counts reports term and language counts
  • no_trace on a translation call, for text that should not be retained
Requirements

A Lara Translate account. The hosted endpoint at https://mcp-v2.laratranslate.com/v1 needs no key at all — the client opens a browser and you log in over OAuth. Running it yourself instead takes an Access Key ID and Secret from the Lara developer console, placed in LARA_ACCESS_KEY_ID and LARA_ACCESS_KEY_SECRET, with npx -y @translated/lara-mcp@latest over stdio or the translatednet/lara-mcp Docker image.

Setup effort

One command plus a key — claude mcp add lara-translate --transport http --url https://mcp-v2.laratranslate.com/v1, then supply credentials