Supervertaler for memoQ

AI translation for memoQ – an LLM machine-translation engine that learns from the segments you confirm, drafts its own project prompt from the document, carries a per-client memory bank with every request, puts your glossary into memoQ's Translation results and into the AI's prompt, and hosts an MCP bridge that lets Claude Desktop translate your live project.

Companion to Supervertaler for Trados, sharing its code through Supervertaler-Plugin-Core.

Status: released, for memoQ 11 and 12, since 28 September 2026. The current version is on the releases page. Download the installer from supervertaler.com/download/memoq; a 14-day trial starts by itself. The add-in is not yet signed by memoQ, so memoQ may ask once whether to load it. What each version changed is in CHANGELOG.md.

Documentation: docs.supervertaler.com/memoq

What it does

AI translation engine. Anthropic, OpenAI or Google, with your own API key and your own instructions. Segment-by-segment while you work, or batched through Pre-translate. Inline tags survive the round trip via memoQ's own segment serialiser. The model list is a short one – the few worth recommending, with a verdict each – and the provider's full catalogue is one tick away. The stable part of every batch request is cached at the provider, and the Activity window shows the token counts that prove it.

Learns from your confirmations. Every segment you confirm is captured and the most relevant ones are shown to the model when it translates later segments in the same document. Confirm electric module once and the rest of the document follows – no configuration, no retraining, just your own approved choices fed forward. Persisted to disk, so it survives closing memoQ.

AutoPrompt. Reads the open document, classifies it, and drafts a project-specific translation prompt with a locked-terms table – confirming the domain with you first and showing exactly what it is about to send. A drafted prompt is the one authority on terminology while it is selected; forbidden glossary terms still travel.

SuperMemory. A memory bank of Markdown articles – brief, terminology, style – per client or per project, sent with every request and in full to AutoPrompt. Remembered per memoQ project; the _shared bank travels whatever else is chosen.

Images. The model sees the text, not the figures. The editor's Images… panel gets the images out of the documents into the memory bank, then has each one described – by the AI, with what the text says about it alongside, or from the text alone for free – and saves the descriptions as figures.md, which every prompt reads. Reference signs seen in a drawing but absent from the text are listed. Documents are found through the live document link; one memoQ cannot place – a project checked out from a server – is located once and remembered.

Terminology. A tab-separated glossary appears as a memoQ terminology provider – matched terms highlighted in the source, entries rendered in the Translation results pane – and the same terms are sent to the model as preferred or forbidden terminology. Forbidden terms are enforced rather than merely displayed. A drafted prompt's locked terms can be exported as a glossary in one click.

Translate with Claude Desktop. The plugin hosts a bridge for the Supervertaler MCP Server, so Claude (or any local-MCP client) can read the document you are translating, your confirmed segments and your glossary, and stage translations that flow into the grid when you press Pre-translate. Tokens go on your Claude subscription, not an API key; every write into your document goes through your own hands. See MCP Server – including the honest table of which Trados tools do and do not exist for memoQ.

A live document link. memoQ's Preview SDK – the interface its own PDF preview uses – is the one channel that shows a tool the target text, the row the cursor is on and the document's real name. Supervertaler.MemoQ.Preview.exe registers as a preview tool and forwards that stream to the plugin, so Claude sees the document as it is, can tell which row you are on, and can ask memoQ to jump to a segment. It draws nothing; it is a link, not a preview.

A companion window. memoQ gives an add-in no window of its own, so everything richer than a settings dialog lives in Supervertaler.PromptEditor.exe – launched from memoQ's dialog or pinned to the taskbar. A panel at the top says what memoQ will use for the next segment (project, model, prompt, glossary, memory bank), each changeable there; beneath it, the prompt library, every memory bank with its articles, and every glossary, any of which can be made active with a right-click and opened for editing. The library is the same one the Trados plugin uses, and Claude can draft prompts into it too. Activity (Ctrl+L) shows what the plugin is doing while memoQ's own dialog says only Processing.

How it fits memoQ

memoQ gives an add-in no window of its own and no API into the project, the editor or its TMs and term bases. A plugin is only ever called: asked for a translation, asked for terminology hits, handed a segment the user confirmed. Everything above is built on those three calls – which is why the AI's knowledge of your document is what has passed through the plugin's hands, why translations from Claude are staged rather than written, and why the glossary is a file rather than a memoQ term base. CLAUDE.md records what the SDK does and does not allow, including a number of things that fail silently.

Installing

memoQ has no plugin marketplace. Copy the DLLs (and the editor) into memoQ's Addins folder – inside the memoQ program directory, so this needs administrator rights:

Supervertaler.MemoQ.dll
Supervertaler.MemoQ.Terms.dll
Supervertaler.PromptEditor.exe

memoQ may warn once that the plugin is unsigned.

Then:

• MT engine – Resource console → MT settings → edit → Services → enable Supervertaler → Configure plugin for provider, model, API key and prompt. To have it learn from confirmations, also set it under Settings → Self-learning MT. API keys live in C:\Users\<you>\Supervertaler\settings\api-keys.json, shared with Supervertaler for Trados and Sidekick, so a key set in any of them works here.
• Terminology – Options → Terminology plugins → tick Perform terminology plugin lookups while working in the translation grid → Supervertaler terms → Options → choose a glossary → Enable plugin.
• Claude Desktop – install Supervertaler-for-memoQ-MCP-Server.mcpb (Settings → Extensions → Advanced settings → Install extension…). It is the same MCP server exe as the Trados extension with SUPERVERTALER_HOST=memoq set; build it with python tools/build_mcpb.py. Other MCP clients: run the exe with that variable set. Steps in the docs.
• Live document link – run Supervertaler.MemoQ.Preview.exe once (the deploy puts it in your Supervertaler data folder, C:\Users\<you>\Supervertaler\memoq\preview\) and accept memoQ's Preview tool connection request, leaving Auto-start with memoQ ticked. memoQ starts it itself from then on.

Glossary format

Tab-separated, one term per line. See examples/glossary-example.txt.

elektrische module	electric module
elektrische module	electrical module	forbidden

A third column containing forbidden marks a target that must not be used. Lines starting with # are ignored; an optional first line #! source=dut target=eng declares the language pair, and a glossary facing the wrong way for the project is reported. The file is re-read whenever you save it, so you can edit it with memoQ open – in any text editor, or as a grid in the companion window, which keeps the header and comments where they were.

Building

Requires the .NET SDK, an installed memoQ (the build references memoQ's own assemblies; nothing is redistributed), and the core/ submodule (git submodule update --init).

bash build.sh              # build, verify, deploy to the Addins folder
bash build.sh --no-deploy  # build and verify only

build.sh refuses to run while memoQ or the prompt editor is open – each locks its own file – and runs tools/smoketest.ps1, which loads the build through memoQ's own add-in loader before deploying. That check exists because memoQ's failure mode is silent: a plugin it cannot load simply never appears, with no error anywhere. After deploying it verifies that what reached Addins is what was built, and says so by name when it is not.

The behaviour harnesses in tools/*-test.ps1 run through tools/run-harness.ps1, which snapshots and restores the shared settings and never lets a test reach a real API key. They must not run while memoQ is open.

Layout

src/Supervertaler.MemoQ	MT engine, options dialog, MCP bridge, capture and staging stores
src/Supervertaler.MemoQ.Terms	Terminology provider (its own DLL: memoQ loads one module per assembly)
src/Supervertaler.PromptEditor	Standalone prompt library editor
src/Supervertaler.MemoQ.Preview	Preview-SDK tool: the live document link (deploys into the Supervertaler data folder, never into Addins)
core/	Shared Supervertaler code (submodule)
tools/	Smoke test, deploy script, the harness suite and its runner, glossary converter, .mcpb builder

Licence

Copyright © 2026 Michael Beijer. All rights reserved.

The source is published so users and reviewers can see what the plugin does. It is not open source: see LICENSE.

---

supervertaler.com · docs.supervertaler.com