# Inglish for AI — local setup

Updated 17 September 2026. The supported approach is the Inglish skill plus the local Inglish+ conversion library. Website conversion APIs and remote MCP conversion are retired. No network conversion fallback is supported.

## Requirements

- A local Inglish+ distribution containing `mcp-server/index.js`, its dependencies, the shared core bundle and `data/library`.
- Node.js and an AI client that can launch a local stdio MCP server and load a skill. Do not assume a mobile app or a cloud chat can launch this desktop setup.
- The Inglish skill (`inglish-skill.zip` or `inglish-SKILL.md`). The ZIP contains `inglish/SKILL.md`; it does not contain the library.

In a source checkout, install the MCP server dependencies from `mcp-server/package.json` using the supplied lockfile if they are absent. In an installed distribution, first verify that the listed server and data files are actually included. An installed app alone is not evidence that its current release packages them.

## Connect the local library

Register a local stdio MCP server named `inglish-plus` with your AI client. Set the command to your installed Node executable and the argument to the **absolute path** of Inglish+'s `mcp-server/index.js`. Consult that client's current local-MCP settings; menus and client support differ. Do not enter a website connector URL.

```json
{
  "mcpServers": {
    "inglish-plus": {
      "command": "node",
      "args": ["/absolute/path/to/Inglish+/mcp-server/index.js"]
    }
  }
}
```

Replace the example path. The server locates `data/library` relative to its package, not the AI client's working directory. Its tools include `convert_to_inglish`, `normalize_to_english` and `list_oov_words`.

## Add the skill

Use your client's supported skill-import mechanism, or extract `inglish/SKILL.md` into its skills directory. Keep the skill's name `inglish`. Client capabilities vary: a cloud-only skill cannot acquire local filesystem or MCP access just by containing these instructions. Use a local-capable client instead of exposing the library publicly.

Ask the AI to list its local tools, convert a short sentence through `inglish-plus`, and confirm the result identifies `inglish-plus-local`. Then request Inglish mode. Confirm code, paths and URLs remain literal. If tools or data are unavailable, the correct result is an explicit setup error in English, never invented Inglish or a website query.

The companion English skill exits the mode. Updating the skill does not update the corpus; keep the Inglish+ library and shared core together from a matching distribution.

## Scope

This local stdio setup targets Windows, macOS and Linux hosts with Node and a compatible AI client. It does not claim native iOS/Android or cloud-only support. Text stays local to the conversion process; your AI client's own model/data policy is separate.