The Python binding is firecrawl-anydoc. It releases the GIL during conversion, so it's safe to mix with asyncio and threads without stalling your whole process.

Install

pip install firecrawl-anydoc

Python stubs are included — type hints work out of the box.

The three core APIs

import anydoc

# 1. From a path
markdown = anydoc.to_markdown("report.docx")
print(markdown)

# 2. From bytes
data = open("slides.pptx", "rb").read()
md = anydoc.to_markdown_bytes(data)

# 3. Stop at the document model (keep embedded assets)
doc = anydoc.to_document(data)
# doc exposes raw bytes and media types of images/objects

Error handling

try:
    anydoc.to_markdown("scanned.pdf")
except anydoc.UnsupportedError as e:
    print("Unsupported format or scanned PDF:", e)

Errors surface as exception subclasses: UnsupportedError, MalformedError, EncryptedError, ResourceLimitError, MissingPartError, IoError. See Error Handling & Limits.

Common scenarios

Concurrent conversion in an asyncio service

import asyncio
import anydoc

async def convert(path: str) -> str:
    return await asyncio.to_thread(anydoc.to_markdown, path)

async def main():
    files = ["a.docx", "b.pdf", "c.xlsx"]
    results = await asyncio.gather(*(convert(f) for f in files))

Because the GIL is released during conversion, to_thread calls don't drag each other down.

Data-analysis pipelines

import anydoc

md = anydoc.to_markdown("sales.xlsx")
# The Markdown tables can go straight into an LLM for analysis

Batch-cleaning contract text

from pathlib import Path

for p in Path("contracts").glob("*.docx"):
    md = anydoc.to_markdown(str(p))
    Path(f"markdown/{p.stem}.md").write_text(md, encoding="utf-8")

Tips

• to_markdown_bytes accepts bytes — file streams and HTTP response bodies work directly
• Want embedded assets? Stop at to_document instead of converting to a string
• A ResourceLimit error on huge files is the safety limit working as intended, not a performance problem

Contributors: souvc