anydoc

anydoc

Fast Rust library that converts documents (Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, and PDF) into clean GitHub-Flavored Markdown. Includes bindings for Node.js, Python, and the browser (WebAssembly).

Built by Firecrawl to turn any office document into LLM-ready Markdown in single-digit milliseconds, with one consistent output no matter which format goes in. It powers Firecrawl Parse, so if you'd rather not run it yourself, the hosted API gives you the same conversion plus our OCR models for the scanned pages anydoc can't read on its own.

Try it in your browser: the demo page runs the library as WebAssembly, so files are converted locally and never leave your machine.

Quick start

Agent skill

anydoc ships as an Agent Skill, so your agent can read any document it runs into:

npx skills add firecrawl/anydoc

The skill teaches the agent to convert documents with the anydoc CLI. Works with Claude Code, Codex, Cursor, OpenCode, and any other compatible agent.

CLI

npx @firecrawl/anydoc report.docx               # Markdown to stdout
npx @firecrawl/anydoc slides.pptx -o slides.md  # or to a file
npx @firecrawl/anydoc - --format csv < data.csv # read stdin
npx @firecrawl/anydoc scan.pdf --ocr hosted     # scanned pages via Firecrawl Parse

npx downloads the prebuilt binary for your platform on first run. For a permanent anydoc command, install globally with npm install -g @firecrawl/anydoc. Run anydoc --help for all options.

Node.js

npm install @firecrawl/anydoc

import { toDocument, toMarkdown, toMarkdownBytes } from '@firecrawl/anydoc';

// From a file path:
const markdown = await toMarkdown('report.docx');

// To enable OCR:
const markdown = await toMarkdown('report.docx', { ocr: 'hosted' );

// From bytes, with the format detected from the content:
const fromBytes = await toMarkdownBytes(bytes);

// Or name it, which signature-less formats (CSV) need:
const fromCsv = await toMarkdownBytes(bytes, 'csv');

// Or stop at the document model, which also carries embedded assets:
const document = await toDocument(bytes);

Full API reference: node/README.md

Python

pip install firecrawl-anydoc

import anydoc

# From a file path:
markdown = anydoc.to_markdown("report.docx")

# To enable OCR:
markdown = anydoc.to_markdown("report.docx", ocr="hosted")

# From bytes, with the format detected from the content:
markdown = anydoc.to_markdown_bytes(data)

# Or name it, which signature-less formats (CSV) need:
markdown = anydoc.to_markdown_bytes(data, "csv")

# Or stop at the document model, which also carries embedded assets:
document = anydoc.to_document(data)

Full API reference: python/README.md

Browser (WebAssembly)

npm install @firecrawl/anydoc-wasm

import init, { toMarkdownBytes, toDocument } from '@firecrawl/anydoc-wasm';

await init();

// From bytes, with the format detected from the content:
const markdown = toMarkdownBytes(bytes);

// Or name it, which signature-less formats (CSV) need:
const fromCsv = toMarkdownBytes(bytes, 'csv');

// Or stop at the document model, which also carries embedded assets:
const document = toDocument(bytes);

Full API reference: wasm/README.md

Rust

cargo add anydoc

// From a file path:
let markdown = anydoc::to_markdown("report.docx")?;

// From bytes, with the format detected from the content:
let markdown = anydoc::to_markdown_bytes(&bytes, None)?;

// Or name it, which signature-less formats (CSV) need:
let markdown = anydoc::to_markdown_bytes(&bytes, anydoc::Format::Csv)?;

// Or stop at the document model, which also carries embedded assets:
let document = anydoc::to_document(&bytes, None)?;

OCR

anydoc reads text-based PDFs locally but does no OCR, so a PDF with scanned or image-only pages fails with NeedsOcr. Opt in and those documents go to Firecrawl Parse, which OCRs them and returns the same Markdown. No signup needed; set FIRECRAWL_API_KEY for higher limits.

	Opt in	Key, else FIRECRAWL_API_KEY
CLI	anydoc scan.pdf --ocr hosted	--api-key <key>
Node	toMarkdown('scan.pdf', { ocr: 'hosted' })	apiKey
Python	anydoc.to_markdown("scan.pdf", ocr="hosted")	api_key

Only documents that need OCR leave the machine, and the whole document goes, since Parse has no page selection. If Parse cannot convert it, Node rejects with code: 'hosted' and Python raises HostedError. --api-url, apiUrl and api_url, else FIRECRAWL_API_URL, point at another Parse deployment. The Rust crate has no ocr option and never makes network calls.

Features

• One output for every format. Each format parses into a shared document model and renders through a single Markdown serializer, so escaping, tables, heading anchors, and footnotes behave identically whether the input was a .doc from 2003 or a .pptx from yesterday.
• Full document structure. Headings with anchors, bold/italic/strikethrough, inline code and code blocks, links and internal cross-references, bulleted/numbered/nested/task lists with the source's own numbering, tables with merged cells and header rows, block quotes, footnotes and endnotes, and speaker notes.
• Equations as LaTeX. Word and PowerPoint (OMML), OpenDocument and EPUB (MathML), and RTF equations convert to GitHub-flavored math: $...$ inline and $$ blocks.
• Embedded assets. Images and embedded objects render as their alt text in the Markdown, and the raw bytes stay available on the document model, tagged with their media type. Images with an external URL become ordinary Markdown images.
• Content-based format detection. The format is read from the bytes themselves (PDF header, RTF open group, OLE stream names, ZIP package mimetype), so mislabeled files still convert correctly.
• Fast. Pure Rust, no ML models, no external services. Median conversion time is under 5ms per document.
• Bindings that stay out of the way. Node.js conversion runs on the libuv thread pool and never blocks the event loop; Python releases the GIL so other threads keep running. TypeScript types and Python stubs ship with the packages.
• PDF support built in. Text-based PDFs convert locally through pdf-inspector, no OCR service required. Scanned pages can opt into hosted OCR.
• Agent ready. Ships as an Agent Skill: one npx skills add firecrawl/anydoc and any agent can read office documents.

Supported formats

Format	Extensions
Word	.doc, .docx, .docm
PowerPoint	.ppt, .pps, .pot, .pptx, .pptm, .ppsx, .ppsm
Excel	.xls, .xlsx, .xlsm, .xlsb
OpenDocument	.odt, .ods, .odp
Rich Text Format	.rtf
EPUB	.epub
CSV	.csv
PDF	.pdf

Benchmark

anydoc is measured against six other converters on 100 real-world documents spanning fourteen formats. Scores run from 0 to 100, higher is better; speed is the median time to convert one document.

tool	formats	median ms	docs judged	score	completeness	structure	formatting	cleanliness
anydoc	14/14	4.4	94	81	87	79	78	81
libreoffice	12/14	1129.5	87	40	59	42	40	24
unstructured	8/14	572.9	58	63	76	59	51	63
markitdown	6/14	134.8	33	65	78	66	60	52
pandoc	5/14	102.1	34	56	74	57	56	38
docling	4/14	513.6	21	57	60	60	57	51
mammoth	1/14	52.5	8	70	84	71	75	51

Per format, like for like:

format	anydoc	libreoffice	unstructured	markitdown	pandoc	docling	mammoth
doc	87	57	67	-	-	-	-
docm	84	48	-	-	-	-	-
docx	88	56	53	71	68	71	70
epub	77	-	72	72	52	-	-
odp	86	23	-	-	-	-	-
ods	82	38	-	-	-	-	-
odt	80	51	68	-	60	-	-
ppt	80	26	-	-	-	-	-
pptx	74	24	-	66	-	52	-
rtf	88	53	46	-	45	-	-
xls	80	38	66	62	-	-	-
xlsm	76	32	-	-	-	-	-
xlsx	72	30	66	55	-	47	-

How quality was scored: an LLM judge (Claude Sonnet 5) compares two tools' outputs blind against ground truth: the document's first six pages, rendered to images by LibreOffice. Each output is scored on completeness, structure, formatting, and cleanliness. Every pair is judged twice with the outputs swapped to cancel position bias, for 482 verdicts in total. Each tool's score averages its per-format scores over the formats it supports, so a corpus heavy in one format can't skew it. It also means each row averages a different set of formats (mammoth's 69 is docx alone, while anydoc's 81 spans all fourteen), so the per-format table is the fair comparison.

Speed is one warm conversion per document on a Ryzen 9 9950X3D (Windows 11, 64 GB DDR5-6400). anydoc and the Python libraries are timed with process spawn excluded; the CLI tools include it, since that is how they are used. The harness lives in bench/; the corpus is not redistributable and is not in the repo.

Best fit: pipelines that receive a mixed bag of office documents and need one consistent, structured Markdown output. In this comparison, anydoc was the only tool to cover all fourteen formats, scored highest on every judged format, and converted documents an order of magnitude faster than the next-fastest tool.

Format detection

The format is read from the file content, using the marker its specification designates: the PDF header, the RTF open group, OLE stream names, the ZIP package mimetype and content types. CSV has no such marker, so the extension or an explicit format names it instead.

Format::from_bytes(&bytes); // Some(Format::Docx), or None when nothing matches
Format::from_extension("pptm"); // Some(Format::Pptx)
Format::from_path(Path::new("report.odt")); // Some(Format::Odt)

The same three functions exist in Node (formatFromBytes, ...) and Python (anydoc.format_from_bytes, ...).

Errors

A conversion returns Err only when no complete Markdown could come out of the file. ConvertError names what went wrong:

match anydoc::to_markdown(path) {
    Ok(markdown) => Some(markdown),
    // No document comes out of these, so record the file and take the next one.
    Err(
        error @ (ConvertError::Encrypted
        | ConvertError::Unsupported(_)
        | ConvertError::NeedsOcr { .. }),
    ) => {
        unconverted.push((path, error));
        None
    }
    Err(error) => return Err(error),
}

Variant	Meaning
Unsupported	Unknown format, or one that cannot be converted
NeedsOcr	Scanned or image-only pages of a PDF, listed in pages
Malformed	Structurally unusable: no meaningful content could be extracted
Encrypted	Encrypted or password-protected
ResourceLimit	Crossed a fixed safety limit (decompression, nesting, node count)
MissingPart	A part required for any meaningful output is absent
Io	The file could not be read, from to_markdown only

Node and wasm publish the variant name on error.code; Python raises one anydoc.ConvertError subclass per variant, or OSError when the file cannot be read.

anydoc converts locally and does not do OCR, so a PDF with scanned pages fails with NeedsOcr. Opt in with ocr: 'hosted' in Node, ocr="hosted" in Python or --ocr hosted on the CLI to send that document to Firecrawl Parse. No signup needed. Set FIRECRAWL_API_KEY for higher limits.

How it works

document bytes
  │
  ├─► format detection      → content markers, not the extension
  │
  ├─► format parser          → one per format (doc, docx, ppt, pptx, xls,
  │                            xlsx, odt/ods/odp, rtf, epub, csv)
  │         │
  │         └─► Document     → shared model: blocks, inlines, tables,
  │                            footnotes, assets
  │               │
  │               └─► GFM serializer → Markdown
  │
  └─► PDF → pdf-inspector    → Markdown directly

Because every format funnels through the same document model and serializer, output quirks get fixed once. A table-escaping fix for docx is automatically a table-escaping fix for rtf, odt, and everything else.

Development

cargo test
cd node && npm install && npm run build && npm test
cd python && pip install maturin && maturin develop && python -m unittest discover -s tests
wasm-pack build wasm --release --target web --scope firecrawl && node --test wasm/test.mjs  # see wasm/README.md

A committed fixture corpus under tests/fixtures/ is snapshot-tested, tests/robustness.rs mutation-tests every fixture, and fuzz/ carries cargo-fuzz targets per format. The speed and quality benchmark lives in bench/.

Releases are tagged v<version>, which publishes the crate, the npm package, and the PyPI wheels from .github/workflows/release.yml. The version lives in three places, bumped together for a release:

• Cargo.toml: the crate
• node/package.json: the npm package
• python/Cargo.toml: the wheel (python/pyproject.toml reads it)

License

MIT