DeepL
by DeepL
Translate text and documents, rephrase and correct writing, and apply your organization's glossaries from your AI assistant. 16 tools, OAuth sign-in, and a documented no-retention commitment on paid plans.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect DeepL via MCP
https://mcp.deepl.com/v1/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
DeepL Tools & Capabilities (14)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Frequently asked questions
Does DeepL store the text I translate through the MCP server?
No, not on a paid plan. DeepL's privacy policy states that texts and documents submitted to DeepL Translator Pro, DeepL API Pro and DeepL Write Pro are not permanently stored, are kept only as long as producing and transmitting the result requires, and are deleted afterwards. The connector requires an active seat-based subscription, so this paid-tier commitment is the one that applies.
Is my text used to train DeepL's translation models?
No on paid plans, yes on the free ones. DeepL's privacy policy states that on Pro services your texts are not used to improve the quality of its services. The same policy says the opposite about the free DeepL Translator and Write: content uploaded there is processed for a limited period to train and improve DeepL's neural networks and algorithms.
What data does DeepL keep when the connector translates a document?
Metadata for 14 days, on paid and free plans alike. DeepL's privacy policy lists what is stored per document translation: status, file type, language pair, time estimates, a reference to the applied glossary, the character count and any errors. It states this database is accessible only to selected employees and is deleted automatically after 14 days.
Can the DeepL MCP server create or edit my glossaries?
No. All six glossary, style-rule and custom-instruction tools are reads. DeepL's REST API does support full glossary management through POST, PATCH, PUT and DELETE calls on its v3 endpoints, so the capability exists in the product and was deliberately left out of the connector. An agent can apply your terminology but cannot rewrite it.
How many characters does a document translation cost?
At least 50,000, whatever the file contains. DeepL's documentation states that every submitted PowerPoint, Word, Excel or PDF document is billed a minimum of 50,000 characters on API plans. All content in the file counts toward the billed total, including text that was not actually translated, so a short slide deck costs the same as a long one.
Can the connector tell me how much of my character allowance is left?
No. None of the 16 tools reports quota. DeepL exposes usage through a REST endpoint, an account dashboard, CSV exports and Admin API analytics, and none of those is wired into the MCP server. An agent translating on your behalf therefore cannot see its own remaining allowance, and you check consumption outside the conversation.
Do rephrase-text and correct-text need a separate DeepL plan?
They need a paid one. DeepL documents its Write API as unavailable on API Free plans and requiring an API Pro subscription. Write also covers far fewer languages than translation: 14 of the 124 language entries in DeepL's own table support text improvement, against all 124 for translation.
How many languages does the DeepL MCP server support?
124 language entries, of which 114 are distinct languages and 10 are regional variants. Every entry supports translation, 43 support glossaries and 33 support style rules. The get-source-languages and get-target-languages tools exist so a client reads current coverage at runtime rather than hard-coding a list that DeepL keeps extending.
Sources
- Live auth posture check against
https://mcp.deepl.com/v1/mcp— an anonymousGETreturns - DeepL Remote MCP (HTTP 200, no redirect, retrieved 2026-08-18). Source for the 16-tool reference table and its Read/Write column, the seat-based subscription requirement, the OAuth 2.1 statement, the out-of-band document flow, and the data handling section. Note: this page is absent from both
developers.deepl.com/llms.txtand the 132-URLsitemap.xml, and is reachable only because Anthropic's directory names it — the documentation index instead lists a different, locally-run MCP server. · retrieved 2026-08-18 - DeepL MCP Server (the local, open-source one) (HTTP 200, retrieved 2026-08-18). Source for the three-server distinction, the eight-tool local list, the
npxinstall and theDEEPL_API_KEYenvironment variable. · retrieved 2026-08-18 - DeepL Privacy Policy (HTTP 200 with a browser user agent, retrieved 2026-08-18). Section 4 is the source for the Pro no-retention and no-training commitments; section 3 for the opposite treatment of the free tiers; section 7.2 for the 14-day document-metadata retention and its per-field list. Section 7.1 also states that document translation is processed on DeepL's own infrastructure in the EEA. · retrieved 2026-08-18
- Usage and limits (HTTP 200, retrieved 2026-08-18). Source for the per-format upload table, the 500,000-character API Free monthly cap, the 128 KiB request-size limit, and the Unicode code point billing unit. · retrieved 2026-08-18
- Document translations (HTTP 200, retrieved 2026-08-18). Source for the 50,000-character document billing minimum, the format-specific caveats, the same-language rejection, the brief storage window and the HTTP 429 and 456 conditions. · retrieved 2026-08-18
- Languages supported (HTTP 200, retrieved 2026-08-18). The counts here were derived by parsing the page's own language table: 124 entries, 114 with
isVariant: false, 124 supporting translation, 43 supporting glossaries, 33 supporting style rules and 14 supporting text improvement. · retrieved 2026-08-18 - Write Quickstart (HTTP 200, retrieved 2026-08-18). Source for Write requiring an API Pro subscription and for the same-language constraint on improvement requests. · retrieved 2026-08-18
- Using Style Rules (HTTP 200, retrieved 2026-08-18). Source for Style Rules being Pro-only and for the create, update and delete routes that the connector does not expose. · retrieved 2026-08-18
- Managing Glossaries (HTTP 200, retrieved 2026-08-18). Source for the full REST glossary lifecycle, which establishes that the connector's read-only glossary surface is a deliberate narrowing. · retrieved 2026-08-18
- Retrieving Usage Data (HTTP 200, retrieved 2026-08-18). Read to establish the absence of a quota tool: DeepL documents four usage-reporting channels — an account UI tab, CSV export, a near-real-time REST endpoint and Admin API analytics — and none is an MCP tool. · retrieved 2026-08-18
- DeepL enterprise security page (HTTP 200 with a browser user agent, retrieved 2026-08-18). Read for corroboration of the retention position. Its marketing claim that texts are never stored or used for training "without your consent" is weaker and vaguer than the privacy policy's tier-specific language, so the privacy policy is cited above in preference. · retrieved 2026-08-18
- DeepL pricing (HTTP 200, retrieved 2026-08-18). Fetched specifically to source plan prices, and it yielded none: the plan table is rendered client-side and the served HTML contains no figures. No price is stated on this page as a result. · retrieved 2026-08-18
robots.txtfordevelopers.deepl.com(HTTP 200, retrieved 2026-08-18). It carriesContent-Signal: ai-train=yes, search=yes, ai-input=yes— a publisher opting in on all three axes, so synthesis of this host's documentation is expressly permitted.www.deepl.com/robots.txt(HTTP 200) blocks only campaign, legacy-API-docs and Cloudflare paths, carries noContent-Signal, and permits the privacy and security pages read here.www.deepl.comserves nollms.txtand nositemap.xml(both HTTP 404). · retrieved 2026-08-18- Anthropic Connectors Directory entry, including the 14 tool names, tier and categories (from the committed directory snapshot of 2026-08-16, re-read 2026-08-18). The entry publishes no
permissionsfield, so this page makes no claim about one. · retrieved 2026-08-16 - DeepL support — · Privacy
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Productivity
- Developer
- DeepL
- Tools
- 14
- Domain
- mcp.deepl.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.