aksara-mcp

MCP server that gives AI agents bidirectional transliteration between Latin-script Javanese and Aksara Jawa (Hanacaraka), powered by aksara-ts.

lamun sira nginguk ucing   →  ꦭꦩꦸꦤ꧀ꦱꦶꦫꦔꦶꦔꦸꦏꦸꦕꦶꦁ
ꦲꦤꦕꦫꦏ                  →  hanacaraka

Aksara Jawa is essentially absent from LLM training data. This server lets agents read manuscript OCR output, transliterate user input, and work with Javanese script in both directions without guessing at rare Unicode codepoints.

Tools

to_aksara

from_aksara

Decoding handles murda consonants, retroflex letters (ṭ, ḍ), cakra (medial r), pengkal (medial y), and standalone vowel letters.

Related MCP server: BPS MCP Server

Install

npm install aksara-mcp

Or clone and build locally:

git clone https://github.com/thesimonharms/aksara-mcp.git
cd aksara-mcp
npm install
npm run build

Requires Node.js 18+.

MCP configuration

Cursor / Claude Desktop

{
  "mcpServers": {
    "aksara": {
      "command": "node",
      "args": ["/absolute/path/to/aksara-mcp/dist/index.js"]
    }
  }
}

If installed globally or via npx:

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

Examples

Latin → Aksara

{ "text": "hanacaraka" }

→ ꦲꦤꦕꦫꦏ

{ "text": "aji saka", "spaces": true }

→ ꦲꦗꦶ ꦱꦏ

{ "text": "aksara", "explicit_vowels": true }

→ ꦄꦏ꧀ꦱꦫ

Aksara → Latin

{ "text": "ꦲꦤꦕꦫꦏ" }

→ hanacaraka

{ "text": "ꦧꦸꦟ꧀ꦝꦼꦭ꧀" }

→ bunḍel

Development

npm run build   # bundle server to dist/
npm start       # run on stdio
npm test        # build + run cobasaja tests

Tests live in tests/ and use cobasaja to spawn the server over stdio and assert tool behaviour end-to-end.

Known limitations

Inherited from aksara-ts:

• ꦲ ambiguity — the glyph is both consonant h and the carrier for standalone vowels. from_aksara on ꦲꦗꦶ returns haji, not aji. Use explicit_vowels: true when encoding if disambiguation matters.

• Spaces — Aksara Jawa traditionally omits word boundaries. Pass spaces: true to to_aksara if you need spaces preserved for round-tripping.

License

MIT © Simon Harms