Runemic OCR, local MCP server

Reads document pages on your computer with Runemic: page images (PNG, JPEG, WebP, GIF; HEIC, TIFF and large scans are converted), PDFs and whole folders. Text comes back in reading order as Markdown or plain text, in any script: Arabic, Persian, Urdu, Hindi, Cyrillic, Chinese, Japanese and more.

Needs Node 18+ and a Runemic API key from console.runemic.com (2,500 free credits). For PDFs, install Poppler (pdftoppm).

Claude Desktop

Download runemic.mcpb and double-click it, then paste your API key.

Any MCP client (Claude Code, Codex, Cursor, VS Code …)

{
  "mcpServers": {
    "runemic": {
      "command": "npx",
      "args": ["-y", "--package=https://runemic.com/mcp/runemic-mcp.tgz", "runemic-mcp"],
      "env": { "RUNEMIC_API_KEY": "rk_live_…" }
    }
  }
}

Tools: read_file (an image or PDF), read_folder (all pages in a folder, in order), list_models, get_account. Pages are sent to the Runemic API to be read and are not stored. Docs: runemic.com/docs#mcp-local.

Prefer no install? The hosted MCP server at https://mcp.runemic.com/mcp needs no key: you sign in in the browser.

Build

npm install && npm run build   # out/runemic-mcp.tgz, out/runemic.mcpb

npm test   # the built server against a stand-in API (tests/mock-api.mjs); needs ImageMagick and Poppler

Every push runs the tests. Releases: bump version in package.json on main; the release workflow tests, builds and publishes both files.

Licence

Apache-2.0