Lokalise MCP Server

The Lokalise MCP Server connects the localization platform Lokalise to AI assistants. MCP, the Model Context Protocol, is an open standard that lets assistants such as Claude, Claude Code, Cursor or Windsurf reach external services. The server comes from Lokalise itself and runs in the vendor's cloud at mcp.lokalise.com. You install no package and host no server of your own. According to the documentation, the feature is currently in beta. On its product page, Lokalise says most teams finish setup in under five minutes.

What is the Lokalise MCP Server good for?

If you ship an app or website in several languages, you maintain hundreds or thousands of translation keys. A translation key is a fixed name such as checkout.button.pay that holds the text in every language. Until now, developers jumped back and forth between their code editor and the Lokalise interface. With the MCP server, the assistant handles that straight from the conversation. A developer in Cursor can ask it to add every new string of a component as a key and attach a screenshot. A localization manager can ask Claude how far along the French translation of a project is, then hand the translators a task with a due date right away.

Two toolkits: project management and software development

Lokalise splits the server into two toolkits, each with its own address. You can connect one or both. The project management toolkit is aimed at localization leads. According to the documentation, it creates projects, adds target languages and contributors, manages glossary terms, uploads files and reports key statistics. It creates and updates tasks, including automatic translation, lists teams and their members, and places translation orders. The software development toolkit is built for developers. It adds keys one at a time or in bulk, updates them and checks translation progress. It also downloads translated language files and manages screenshots. Screenshots give translators visual context, so a short word like Save gets translated in the right sense.

Sign-in and permissions

Lokalise recommends OAuth 2.0. OAuth is a sign-in method where you confirm access in your browser and keep no password or token in your configuration. An API token works as well. You create it in your profile settings and send it in the apikey header. A read-only token is enough for lookups. Creating, updating and uploading require a token with read and write rights. One point matters most: the assistant can never do more than the account you sign in with. Your existing Lokalise roles and project permissions stay in force. Because the server sits on top of the Lokalise API, the API's rate limits apply too.

How does it differ from alternatives?

There are several unofficial community Lokalise MCP servers on npm and GitHub. They run locally, usually need an API token and are not maintained by Lokalise. The official server is hosted, covers far more ground and follows the platform's permission model. Lokalise also stresses that MCP does not replace the REST API. For recurring jobs in scripts or a CI pipeline, the API remains the right tool. The MCP server is meant for interactive work in conversation with the assistant.

Cost and requirements

According to the documentation, the server itself costs nothing extra. You do need a Lokalise account on a plan with API access. Lokalise has no free plan. Instead it offers a 14-day trial with no credit card, followed by several paid plans up to Enterprise. Current terms are on the official pricing page. The documentation states that the server is available in Lokalise Expert but not in Lokalise Vantage. Keep one more thing in mind: automatic translation tasks use Lokalise AI by default. If you explicitly ask for Google or DeepL, those tasks consume processed words from your allowance.

Who is the Lokalise MCP Server for?

The server pays off for product teams that already use Lokalise and spend a lot of time on routine work: adding keys, checking progress, assigning tasks, downloading files. Developers stay in their editor, and localization managers run projects through chat. If you do not use Lokalise yet, you can try the server during the trial. It is not meant for one-off translations without the platform.

A Lokalise account on a plan with API access (Lokalise Expert, not Vantage). Plus an MCP-capable client such as Claude Code, Cursor, Claude, Windsurf or ChatGPT. For write actions via API token, a token with read and write rights.

The Lokalise MCP Server gives your AI assistant access to your localization projects. It runs in the Lokalise cloud, so there is nothing to install. This guide shows which toolkit you need, how to connect it in Claude Code and Cursor, and what to watch out for afterwards.

Step 1: Check first

You need a Lokalise account on a plan with API access. According to the documentation, the server is available in Lokalise Expert, not in Lokalise Vantage. The feature is still in beta. The assistant gets exactly the rights of your account. If you can only read in Lokalise, the assistant cannot change anything either.

Pick the right toolkit

Lokalise offers two toolkits, each with its own address. The project management toolkit at https://mcp.lokalise.com/mcp/project-management is meant for localization leads: projects, languages, contributors, glossary, tasks and translation orders. The software development toolkit at https://mcp.lokalise.com/mcp/software-development is for developers: adding and updating keys, managing screenshots and downloading language files. You can connect both at once. If you only maintain keys, the development toolkit is enough and keeps the assistant's tool list lean.

Step 2: Claude Code

Run claude mcp add lokalise_sd --transport http https://mcp.lokalise.com/mcp/software-development. For the project management toolkit, the command is claude mcp add lokalise_pm --transport http https://mcp.lokalise.com/mcp/project-management. Then restart Claude Code and finish sign-in in your browser. Use claude mcp list to check that both servers are connected.

claude mcp add lokalise_sd --transport http https://mcp.lokalise.com/mcp/software-development
claude mcp add lokalise_pm --transport http https://mcp.lokalise.com/mcp/project-management
claude mcp list

Cursor

Open Cursor Settings, then Tools & MCP. In the MCP configuration section, add an entry lokalise_sd under mcpServers with the url field set to the development address. For the second toolkit, add lokalise_pm with the project management address. After saving, Cursor asks you to sign in.

Step 3: Signing in with an API token instead of OAuth

OAuth is the recommended option because no token sits in your configuration. If your client does not support OAuth, create a token in Lokalise under Profile settings and API tokens. In the configuration entry, set type to http and add the headers apikey with your token and accept with the value application/json, text/event-stream. A read-only token is enough for lookups. To create and update, you need a token with read and write rights.

First steps in daily work

Open a new conversation and explicitly ask the assistant to use the Lokalise MCP server. Good starter tasks are asking for the translation progress of a project or adding new keys from a component you just wrote. Name the project so the assistant does not have to guess. For recurring jobs in your CI pipeline, the REST API remains the better choice.

Work safely

With write access, the assistant can change many keys at once and place translation orders that cost money. Approve such actions one at a time. Automatic translation uses Lokalise AI by default. If you explicitly ask for Google or DeepL, it consumes processed words from your allowance. Also keep an eye on the API rate limits, which apply to the assistant as well.

OAuth 2.0 (recommended), with no token stored in the client configuration. Alternatively a Lokalise API token from your profile settings in the apikey header. Read-only tokens are enough for lookups; creating, updating and uploading need a token with read and write rights.

Access matches exactly the rights of the signed-in account or the token creator, including project access and allowed actions (role-based access control). The project management toolkit can also add contributors, create projects and place translation orders.

The server runs in the Lokalise cloud and uses the Lokalise API. Text and project data the assistant fetches is passed to the respective AI provider and handled under that provider's terms. Automatic translation uses Lokalise AI by default, and Google or DeepL only when explicitly requested.

Medium. With write access, the assistant can change keys in bulk, add contributors and place paid translation orders. A misunderstood instruction or crafted content can trigger unwanted changes. Use a read-only token for pure reporting and approve bulk changes and orders one by one.