Deploy anydoc | (Just Updated) Any Document to Markdown API, Japanese & Thai OCR

Deploy and Host anydoc on Railway

anydoc-serve turns Word, PowerPoint, Excel, OpenDocument, PDF, EPUB, RTF and CSV files into clean Markdown over HTTP and MCP. It wraps firecrawl/anydoc, Firecrawl's fast Rust document converter, in one small container and adds local OCR for scans and images in English, Japanese and Thai.

About Hosting anydoc

This template runs one service from ghcr.io/bon5co/anydoc-serve:latest (amd64 and arm64). There is no database and no volume: every request is converted in memory and nothing is stored. Born-digital documents go through anydoc unchanged; scanned or image-only pages, and images (.png, .jpg, .tiff, .webp), are OCR'd on CPU with Tesseract 5, reading Japanese, Thai and English in one pass, so nothing leaves your deployment. On a mixed PDF only the scanned pages are OCR'd. Every endpoint except /health requires Authorization: Bearer $API_KEY, and the key is generated for you at deploy, so no deployment is ever open. A /health healthcheck keeps traffic off the service until it is ready. Measured on a Railway deploy of this template: 246-263 MiB of RAM idle and 272 MiB after converting a DOCX and OCR'ing two images, under the Free plan's 0.5 GB cap.

Why Deploy anydoc on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying anydoc on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.

• Scans too, not just born-digital files — plain anydoc refuses scanned PDFs and images; this server OCRs them locally instead of sending them to a hosted OCR API.
• Japanese and Thai OCR out of the box — OCR_LANGS defaults to eng,jpn,tha, and one page mixing all three scripts is read line by line.
• Private by default — a 32-character API key is generated per deploy; requests without it get 401.
• Built for agents — an MCP server at /mcp exposes a convert_document tool for Claude Code, Cursor and any client that speaks streamable HTTP.
• Small — one service, no volume, about 250 MiB idle, 272 MiB after OCR.

Common Use Cases

• RAG ingestion — convert uploaded Office files, PDFs and scans to Markdown before chunking and embedding.
• AI agent tools — give Claude Code or Cursor a convert_document tool over MCP so the agent can read any attachment.
• Japanese and Thai back-office documents — OCR scanned invoices, forms and contracts without a cloud OCR bill.

Dependencies for anydoc Hosting

• Nothing beyond the one service in this template: no database, no volume, no external API.

Deployment Dependencies

• bon5co/anydoc-serve — the server this template deploys (source, Dockerfile, OCR benchmark)
• firecrawl/anydoc by Firecrawl — the document converter doing all non-OCR conversion (MIT). anydoc-serve is not affiliated with Firecrawl.
• Tesseract — the OCR engine (Apache 2.0)

Implementation Details

Convert a document to Markdown:

curl -H "Authorization: Bearer $API_KEY" -H "Accept: text/markdown" \
  -F file=@report.docx https://your-deployment.up.railway.app/v1/convert

OCR a scan or an image, getting JSON with the Markdown plus the pages that went through OCR:

curl -H "Authorization: Bearer $API_KEY" \
  -F file=@scan.pdf https://your-deployment.up.railway.app/v1/convert

Convert from a URL:

curl -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/report.pdf"}' https://your-deployment.up.railway.app/v1/convert

Add it to Claude Code as an MCP server:

claude mcp add --transport http anydoc https://your-deployment.up.railway.app/mcp \
  --header "Authorization: Bearer $API_KEY"

More settings (OCR_ENABLED, OCR_DPI, MAX_UPLOAD_MB, CONVERT_TIMEOUT_S, ...) are documented in the anydoc-serve README.