JSON Link Model Context Protocol (MCP) Server

An official Model Context Protocol (MCP) server for JSON Link, allowing AI assistants (Claude Desktop, Cursor, Antigravity, Cline, Windsurf) to directly parse, lint, validate, convert, and export software localization files.

Capabilities & Tools

| Tool | Description | | :--- | :--- | | convert_zawgyi | Lossless Rabbit converter between Myanmar Zawgyi font encoding and international Myanmar Unicode standard with automatic detection. | | validate_variables | Ensures all interpolation variables ({username}, %s, {{count}}, $name) from source strings are preserved in translations without syntax damage. | | lint_translations | Comprehensive QA scanner checking for leading/trailing whitespace, missing keys, variable mismatches, text expansion risk, and Zawgyi font issues. | | read_translations | Parses translation files (.json, .jsonlink, .yaml, .xml, .strings) into normalized JSON Link records. | | export_bundle | Generates framework-ready localization files (Next.js / React JSON, Flutter ARB, iOS Localizable.strings, Android strings.xml, TypeScript d.ts). |

Installation & Setup

1. Zero-Install via NPX (Recommended)

You can run the MCP server directly via npx in any AI IDE or assistant without cloning or manual building:

In Claude Desktop

Add this to your Claude Desktop configuration file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "json-link": {
      "command": "npx",
      "args": ["-y", "@jsonlink/mcp"]
    }
  }
}

In Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "json-link": {
      "command": "npx",
      "args": ["-y", "@jsonlink/mcp"]
    }
  }
}

In Antigravity

Add to ~/.gemini/antigravity/mcp_config.json:

{
  "mcpServers": {
    "json-link": {
      "command": "npx",
      "args": ["-y", "@jsonlink/mcp"]
    }
  }
}

2. Local Development & Build

If developing locally on the server:

cd mcp
npm run build

Example AI Prompts

Once configured, you can prompt your AI assistant naturally:

• "Check if our mobile app's Myanmar translations in locales/my.json contain any Zawgyi font encoding and convert them to standard Unicode."
• "Audit locales/ for missing variables or trailing whitespaces using JSON Link linter."
• "Validate whether {username} and {count} were correctly preserved in this translation."
• "Export our translations into Flutter ARB and iOS Localizable.strings."

Testing via CLI

You can test the MCP server manually using Node.js:

node -e '
const { spawn } = require("child_process");
const proc = spawn("node", ["mcp/dist/index.js"]);

proc.stdout.on("data", (d) => console.log(d.toString()));

proc.stdin.write(JSON.stringify({
  jsonrpc: "2.0",
  id: 1,
  method: "tools/call",
  params: {
    name: "convert_zawgyi",
    arguments: { text: "မဂၤလာပါ" }
  }
}) + "\n");
'

License

MIT © Pyae Phyo Maung