~/guides/ANYDOC.md

Guides / AI tools

AnyDoc: PowerPoint, Excel and Word to Markdown in one command, so Claude Code can read them (3 files tested)

Firecrawl's open-source AnyDoc turns Word, PowerPoint, Excel, PDF and EPUB into clean Markdown. One npx line, a Claude Code skill, real output, the OCR limit.

Hand Claude Code a PowerPoint or Excel file and it cannot read it directly and has to write a script to unpack it first. Firecrawl's open-source AnyDoc turns Word, PowerPoint, Excel, PDF and EPUB into clean Markdown in one command, keeping tables, chart numbers and speaker notes. This is the full guide to the ANYDOC reel: I converted three files for real, so here is why you convert, how, what comes out, how to install it as a skill so Claude Code does it on its own, and the limit with scanned PDFs.

01The result first: the same deck as Markdown Claude reads at a glance

Reel frame: the Q3 report slide on the left, the converted Markdown on the right
Reel frame: the real slide on the left, the Markdown AnyDoc wrote from the same .pptx on the right.
Slide: 小島咖啡 2026 Q3 report, a revenue table with a merged header
Before: slide 1 of our demo deck (小島咖啡 is a made-up coffee chain).
Editor view: the same table as a Markdown table, the speaker notes below as a quote
After: the .md AnyDoc wrote, table and speaker notes included.

When you're done you'll have:

  1. An .md file with the headings, tables, bullets and speaker notes in their original structure.
  2. A Claude Code skill that converts Word, PowerPoint and Excel files on its own before reading them.
  3. A clear idea of which files convert on your machine and which (scanned PDFs) go to the cloud.

02Why Claude Code can't read PowerPoint and Excel

Reel frame: a .pptx cube bursts into dozens of small file cubes; slide1.xml on the wall monitor
One deck, unzipped: our .pptx holds 50 files.

Claude Code's file-reading tool refuses files it treats as binary, such as .xlsx; then it has to write a script or use a skill to unpack them. That is because .pptx, .docx and .xlsx are really zip archives full of XML.

  1. Our two-slide deck unzips into 50 files.
  2. slide1.xml alone is 7,308 characters, mostly layout tags.
  3. The whole deck as Markdown is 969 bytes.

03Step 1: convert with one command

Reel frame: a laptop terminal printing the Markdown after npx @firecrawl/anydoc on the .pptx
Real capture: the run on this computer.

If you have Node.js you can run it straight away; the first run downloads the build for your computer:

Run in a terminal (replace the brackets)
npx @firecrawl/anydoc [report.pptx] -o [report.md]

If you use it a lot, install the command:

Or install it
npm install -g @firecrawl/anydoc
anydoc [report.docx] -o [report.md]
For Python
pip install firecrawl-anydoc

04Step 2: check what came out

What our three files turned into:

  1. The deck's table became a Markdown table; the merged header became an empty cell and every number was right.
  2. The bar chart's numbers became a table, so Claude can read July, August and September.
  3. Speaker notes became quotes starting with ">" after their slide.
  4. Bullets kept their levels; the Excel formula totals (377, 354, 731) came through.
  5. The Word contract's headings, table and 1. 2. 3. numbering stayed in order.

05Step 3: install it as a skill so Claude Code converts on its own

Reel frame: the wall monitor shows the README's Agent skill section with npx skills add firecrawl/anydoc circled in red
Official README: AnyDoc ships as an Agent Skill that works with Claude Code.

If you don't want to convert by hand every time, install it as a skill. In your project's terminal:

Run in a terminal
npx skills add firecrawl/anydoc

Then just hand Claude Code the file:

Tell Claude Code (replace the brackets)
Read [report.pptx], give me three key points, then turn the numbers in its tables into one comparison table.

06Good to know

Reel frame: a paper browser with the README OCR section, --ocr hosted below it and an "uploads the whole file" stamp
Official README: scanned PDFs need OCR, which goes to Firecrawl's cloud.
  1. Scanned or image-only PDFs fail (it stops). Only with --ocr hosted do they go to Firecrawl Parse for OCR, and the whole file is uploaded; no signup, FIRECRAWL_API_KEY raises the limits. Think twice with confidential files.
  2. Don't want to install anything? The official browser version at firecrawl.github.io/anydoc converts on your computer and never uploads the file.
  3. Images keep only their alt text; the pictures themselves are not in the Markdown.
  4. The comparison scores in the README are Firecrawl's own benchmark; treat them as a guide.
  5. AnyDoc is an MIT-licensed open-source project, not an official Anthropic tool.
  6. You found this guide by typing ANYDOC on the site from the link in bio; there is no DM.

Sources

Related guides