What is anydoc?
anydoc is a Rust library built by Firecrawl that converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF files into GitHub-Flavored Markdown. It targets developers processing mixed document collections for LLM input and other text-based workflows. Office formats use a shared document model and Markdown serializer, while text-based PDFs are handled through pdf-inspector.
The tool can run locally through its CLI or Rust, Node.js and Python interfaces, or inside a browser through WebAssembly. Its agent skill teaches compatible agents to use the CLI. Conversion preserves structures such as headings, tables, lists, footnotes and speaker notes, and converts supported equations into LaTeX. Embedded image bytes remain accessible through the document model rather than being embedded in the Markdown output.
anydoc does not perform OCR locally. PDFs containing scanned or image-only pages fail with an error identifying those pages. The CLI, Node.js and Python interfaces can optionally send such documents to Firecrawl Parse for hosted OCR. This transfers the entire document, not just the scanned pages; the Rust crate never makes network calls.
Key capabilities
- Document conversion Converts office files, ebooks, CSV and text-based PDFs into GitHub-Flavored Markdown.
- Structured output Preserves headings, tables, lists, footnotes, cross-references and speaker notes, with supported equations rendered as LaTeX.
- Format detection Identifies supported formats from content markers; CSV requires an extension or an explicit format.
- Embedded assets Exposes image and object bytes through the document model, while embedded images appear as alt text in Markdown.
- Developer interfaces Provides a CLI and Rust, Node.js, Python and WebAssembly interfaces, including TypeScript types and Python stubs.
- Agent access Includes an agent skill compatible with Claude Code, Codex, Cursor and OpenCode.
Where anydoc fits
anydoc fits into document ingestion pipelines that receive office files, spreadsheets, presentations and PDFs and need Markdown for downstream processing. Applications can call the library directly, browser tools can convert document bytes locally, and compatible agents can use the skill to access files they cannot read directly.
Before deployment, teams should test representative documents for structure and asset handling, and implement error handling for encrypted, malformed, unsupported or resource-limited files. Scanned PDF workflows require a separate OCR decision: enabling the hosted fallback transfers the whole document to Firecrawl Parse. Browser deployments should also account for synchronous, single-threaded WebAssembly conversion and use a worker when interface responsiveness matters.
Install
npx skills add firecrawl/anydocReviews of anydoc
Be the first to share your experience with anydoc.
Write a review
Loading review form…