DeepL MCP Server

The DeepL MCP Server brings DeepL translation straight into your AI assistant. MCP, the Model Context Protocol, is an open standard that lets assistants such as Claude, ChatGPT or Cursor reach external services. DeepL offers the server in two forms. The first is a hosted server at mcp.deepl.com/v1/mcp, which you connect to your DeepL account with nothing to install. The second is an open-source server under the MIT license that you run locally with an API key. Both come from DeepL SE in Cologne, the company behind the well-known DeepL Translator.

What is the DeepL MCP Server good for?

AI assistants can translate on their own, but with technical texts, fixed terms and formal writing, their results often vary. The server solves this by turning every translation into a real call to DeepL. According to DeepL, this makes it clear which result actually came from DeepL. You could, for example, have an email translated into formal German and French in a single step. You could translate a contract PDF and keep its tables, headings and images in place. Or you could have a draft rephrased and proofread without leaving the chat. For teams with glossaries, the biggest gain is that product names and technical terms stay the same in every translation.

The hosted remote server

DeepL recommends the hosted server. It uses Streamable HTTP, a transport method for remote MCP servers. Sign-in uses OAuth 2.1. This is a sign-in method where you grant access in your browser without handing over a key. In Claude, ChatGPT and Microsoft Copilot, DeepL is available as a ready-made connector in each directory. Other clients such as Cursor, VS Code or Copilot Studio connect through the server address. The server offers up to 29 tools. They cover translating, rephrasing and proofreading, plus uploading and downloading documents and managing glossaries, style rules and custom instructions. According to DeepL, the server does not keep text beyond the request. Documents are only streamed through and deleted shortly after translation.

The open-source server for local use

For developers, there is the deepl-mcp-server package on GitHub and npm. It runs over stdio, meaning as a local process that the MCP client starts itself. You need Node.js 18 or later and a DeepL API key. The server translates text and local files such as PDF, DOCX, PPTX or XLSX and writes the result to disk. It also rephrases text and reads glossaries and style rules from your account. This version cannot create or change glossaries or style rules. Its strength is that coding agents such as Claude Code can work directly with files in your project.

Free account and paid plans

According to DeepL, the connector itself is free and included in your subscription, with no separate bill. A free DeepL account is enough to get started. It does lack glossaries, style rules, formality, context and custom instructions, though. Its character limits per request are also much lower than on the seat-based plans. Usage is subject to fair-use limits. The local server bills through your API plan. You can find current plans on DeepL's pricing page.

Who is the DeepL MCP Server for?

The server pays off for anyone who works in several languages on a regular basis and already uses DeepL. That includes marketing and support teams, translation agencies and companies with fixed terminology. Glossaries and style rules keep texts consistent, whether the translation happens in a chat or in the editor. Developers who translate interface strings or documentation will get more out of the local server. If you only translate a short paragraph now and then, the assistant alone will usually do. Compared with unofficial community DeepL servers, this offering has a clear edge. DeepL maintains it, it uses OAuth instead of fixed keys, and administrators can control it for the whole organization.

For the hosted server, a DeepL account, either free or on a seat-based plan (Individual, Team, Business, Enterprise), plus a client that supports remote MCP over Streamable HTTP with OAuth. On Team and Enterprise plans, an admin may need to enable DeepL first. For the local server, Node.js 18 or later and a DeepL API key.

The DeepL MCP Server lets your AI assistant translate text and documents with DeepL. It comes hosted at https://mcp.deepl.com/v1/mcp and as an open-source server for local use. This guide covers both routes and what to keep in mind afterwards.

Which version fits?

For most people, the hosted server is the easiest route. You sign in with your DeepL account and need no API key. It can also manage glossaries and style rules. Choose the local server if a coding agent such as Claude Code should translate files in your project directly. For that, you need Node.js 18 or later and a DeepL API key.

Claude, ChatGPT and Microsoft Copilot

DeepL is listed as a ready-made connector in all three assistants. Search for DeepL in the Claude connector directory, under Apps in ChatGPT or in the Microsoft Marketplace and connect it. Sign in with your DeepL account in the browser and approve access. On Team and Enterprise plans, an admin may need to enable DeepL first.

Cursor, VS Code and other clients

Add a new MCP server or custom connector in your client and enter the address https://mcp.deepl.com/v1/mcp. The client discovers the sign-in endpoints by itself, so you need no client ID or secret. Sign in with your DeepL account and approve. If DeepL does not recognize the client, an extra confirmation step follows.

Step 1: Locally in Claude Code

Create an API key in your DeepL account. Then run claude mcp add deepl --env DEEPL_API_KEY=your-key -- npx -y deepl-mcp-server in your terminal. For other clients, add a deepl entry to your MCP config with the command npx, the args -y and deepl-mcp-server, and an env block with DEEPL_API_KEY. Without the key, the server exits right away.

claude mcp add deepl --env DEEPL_API_KEY=your-key -- npx -y deepl-mcp-server

Translating documents in Claude

Document translation needs Claude to have network access to mcp.deepl.com. On Team and Enterprise plans, an owner allows this in the organization settings under the network access features. Without it, the assistant shows an upload widget in the chat. Behind a firewall or proxy, mcp.deepl.com must be allowed as well.

Limits on the free account

The free DeepL account lacks glossaries, style rules, formality and custom instructions. Its character limit per request is also much lower than on the seat-based plans. Each target language counts as a separate translation, even if you name several in one request.

Work safely

The hosted server can change and delete glossaries and style rules. Review these actions before you approve them. The local server writes files to your disk. Keep your API key out of config files you share with others. To end the hosted server's access, revoke its token.

Hosted: OAuth 2.1 with your DeepL account, no API key; on Enterprise with SSO through the organization's identity provider. Access ends immediately once the token is revoked. Local: DeepL API key in the DEEPL_API_KEY environment variable.

Translating, rephrasing and proofreading text, uploading and downloading documents. Depending on plan and team permissions, also creating, changing and deleting glossaries and style rules. The local server reads and writes files on the machine it runs on and only reads glossaries.

According to DeepL, text is not kept beyond the request and is not used to train models. Document sessions are kept briefly and then deleted. Traffic is encrypted with TLS, and the server only talks to DeepL services. Content in the chat also goes to the respective AI provider.

Low to medium. The remote server can change and delete glossaries and style rules, and the local server writes files to disk. A crafted text with hidden instructions can trigger unwanted changes. Review write actions before approving them and keep your API key out of shared configs.