Use DeepL MCP with Sokko.

Correct one or more texts for typos, grammar and punctuation errors using DeepL.

Get a download link for a translated document once its status is 'done'. Returns 'downloadUrl', a short-lived, single-use link. Fetch it with an HTTP GET (the URL carries its own token, so do not add an Authorization header), or present it to the user. The link works only once. The bytes are streamed from the server and never pass through this conversation. For an upload with several target languages: omit 'targetLang' to get one link per finished language, or pass it to get that one language.

Check the translation status of a document session. Returns one of: 'awaiting_upload' (file not received yet), 'queued', 'translating', 'done', or 'error'. The additive 'uploadStatus' is 'awaiting', 'uploading', or 'complete' when known. Any of the upload's session ids reports the whole upload: 'status' covers every target language ('translating' while any language still runs, 'error' only when they all failed) and 'targets' reports each language on its own. Once the status is 'done', call download-document.

Get the source language codes supported by DeepL for translation, e.g. 'EN' or 'DE'. Use one of these for the sourceLang parameter of translate-text.

Get the target language codes supported by DeepL for translation, e.g. 'EN-US' or 'DE'. Use one of these for the targetLang parameter of translate-text.

Rephrase text in the same or a different language using DeepL.

Translate text to one or more target languages using DeepL. Use this for plain text provided directly in the conversation — snippets, strings, messages, or passages pasted by the user. Do not use this to translate a file or document (e.g. Word, PowerPoint, Excel, PDF, HTML, .txt, .srt, .xlf/.xliff) — even if you can already read its contents — because it returns plain text and discards the document's layout and formatting. For files, use upload-document instead. Review the optional parameters and use those that apply for best results. When using a glossary, you must specify the source language as well as the target language. If the user references a glossary by name, use list-glossaries to look up its id. An unknown glossaryId is silently ignored (translation proceeds without it). An administrator may assign a language system (style profile) to the account, whose glossaries and style rule set can replace the glossaryId and styleId passed here; 'appliedCustomizations' in the result says so when it happened, per target language. Every target language is translated in one call; the result holds one entry per language, and a language DeepL refused is reported in 'failures' while the others still translate. Multiple texts are supported and translated independently; within each target language, results correlate to the inputs by position. If any text is blank, the whole request is rejected.

Translate a whole file or document, preserving its original layout and formatting. Use this — not translate-text — whenever the user wants to translate a file or document (rather than text typed into the conversation), even if its contents are already visible to you; translate-text would strip the formatting. For plain text pasted into the conversation, use translate-text instead. Supported file extensions in this deployment: .docx, .doc, .pptx, .ppt, .xlsx, .xls, .pdf, .htm, .html, .txt, .xlf, .xliff, .srt, .png, .jpg, .jpeg. PNG and JPEG files are limited to 3 MB; limits for other formats depend on the user's plan. Always start a document translation with this call, and pass 'filename' with the file's extension. It reserves a short-lived upload session and returns 'sessionId' and 'uploadUrl'; it does not upload the file itself, and nothing is translated or billed until the file arrives. If the call had no 'filename', first call upload-document again with it. Then do exactly one of the following: 1. Upload the source file bytes: HTTP PUT the raw bytes to 'uploadUrl' with 'Content-Type: application/octet-stream'. 2. If you cannot access or transfer the file bytes, immediately call upload-document-show-fallback-widget with the returned 'sessionId'; it shows the user an upload widget where they select the file. Being unable to upload the bytes directly is an expected, recoverable condition: the next step is then to call upload-document-show-fallback-widget. Pass all target languages in one call rather than one call per language: the file is then uploaded only once. Each language is translated and billed separately once the file arrives. A language DeepL refuses is listed in 'failures' while the others still translate. An administrator may assign a language system (style profile) to the account, whose glossaries and style rule set can replace the glossaryId and styleId passed here; 'appliedCustomizations' on a target says so when it happened, per target language.

Second step of a document translation, for when the file bytes cannot be sent to 'uploadUrl': shows the user an upload widget for the upload session that upload-document already reserved, so they can select the file themselves. Call it with the 'sessionId' returned by upload-document right after that call, whenever you cannot access or transfer the file bytes. The widget tracks the translation and offers the download.