← Knowledge base

PelaPela MCP Server

Last updated: 2026-09-06

What it is

PelaPela is a Model Context Protocol (MCP) server that tunes an LLM's translation output for naturalness, tone, and script choice, parameterized by language pair.

LLM translations often read as translated rather than natural: literal transliteration of ordinary words, mismatched formality, or the wrong script for the target audience. PelaPela feeds the model guidance for a specific pair so the output sounds like something a native speaker would actually write.

What it exposes

One prompt and one tool, both named pela_translate_tuning:

Parameter Required Default Notes
pair yes Language pair, e.g. en-jp, en-es
tone no neutral e.g. friendly, formal
script no Where the pair supports one, e.g. kanaonly for en-jp

Supported pairs today: en-jp, en-es.

No API key or authentication is required.

Install

Local (stdio)

npx pelapela-mcp

Client config (Claude Desktop, Claude Code, or any MCP client):

{
  "mcpServers": {
    "pelapela": { "command": "npx", "args": ["pelapela-mcp"] }
  }
}

Hosted (Cloudflare Workers)

Point your client at the remote endpoint instead of spawning a local process:

https://pelapela-mcp.jess-901.workers.dev/mcp

Use it

In Claude Code the tuning is available through skill-style syntax:

/pela:en-jp friendly kanaonly
/pela:en-es formal

In any MCP client, call the pela_translate_tuning tool (or attach the prompt of the same name) before asking the model to translate. The returned guidance becomes part of the model's context for that language pair, tone, and script.

Relationship to the Translation API

The MCP server only shapes how another model translates — it does not translate text itself. For direct machine translation use the Translation API.