The vendor now runs the server itself at `https://mcp.handwritingocr.com`, with OAuth instead of a copied token and a larger set of capabilities — Word, Excel and JSON export, plus field extractors. Point your client at that URL and there is nothing to install. Keep the local server for the cases the hosted one cannot serve: an air-gapped machine, or a client that cannot reach a remote MCP server at all.
An MCP server for the Handwriting OCR service. You upload an image or a PDF, poll its status while the service works, and retrieve the transcription as Markdown — the three steps that a handwriting job actually takes, since transcription is not instantaneous.
- An image or PDF document submitted for transcription
- The status of a submitted document checked while it is processing
- The finished transcription retrieved as Markdown, ready to paste or edit
- A hosted endpoint at `https://mcp.handwritingocr.com` with a wider surface than the local server: the same submit, status and retrieve cycle plus export to Word, Excel or JSON, and extractors that pull named fields out of a form rather than transcribing it whole
A Handwriting OCR account. For the hosted endpoint that is all: authorization runs over OAuth in your browser, with no token to copy. For the local server you also need an API token from the platform's API settings, supplied as `API_TOKEN`, and Node.js 18.x or higher. The npm package name is `handwriting-ocr` (0.1.0) and it is marked private, so build from a clone or use the Smithery installer. Stdio-only clients can still reach the hosted endpoint by bridging through `mcp-remote`.
One command plus a key — npx -y @smithery/cli install @Handwriting-OCR/handwriting-ocr-mcp-server --client claude, then supply credentials