An MCP server, so your coding agent can write the strings too

An agent adding a screen invents copy as it goes. Without somewhere to put it, that copy is hard-coded English and someone finds it in production. The MCP server gives the agent the same six operations a developer has.

Six tools

One file, stdio, no dependencies to install.

`list_platforms`

The apps in the workspace, their default format and their locales.

`list_projects`

Client variants under a platform, for white-label wording.

`pull_translations`

One locale as a file, in the platform's own format.

`push_source_strings`

New English keys, straight from the feature being written.

`translation_status`

What is finished, what needs review, what is missing a plural form.

`list_glossary`

The terms that must not be translated, so the agent does not invent a second word for your product.

Add it to a client

Everything below authenticates with one API key, scoped to read or write and optionally pinned to a single platform, so a build job can never reach further than the app it builds.

claude_desktop_config.json / .mcp.json

{
  "mcpServers": {
    "locoloop": {
      "command": "python3",
      "args": ["/path/to/locoloop/mcp/server.py"],
      "env": {
        "LOCOLOOP_API_KEY": "llk_...",
        "LOCOLOOP_BASE_URL": "https://api.rondigi.com"
      }
    }
  }
}

Why it is safe to hand over

The key decides, not the agent

Scope it to read, or pin it to one platform. The server has no notion of who is calling — the API enforces it.

Nothing is overwritten

push_source_strings flags existing translations for review. A human still approves the change.

Every edit is attributed

Agent writes land in the same history as human ones, with a revert on every cell.

Point your agent at a real workspace

Create one, mint a key, paste the config. It takes about five minutes.

Start free Log in Book a demo