PDF skill

The PDF skill gives Claude a toolbox and a playbook for almost anything you can do to a PDF: read it, pull tables out, combine files, stamp a watermark, run text recognition on a scan, build a new PDF, or fill in a form. You get files back, not just a description of them.

What it does (plain English)

Unlike the Word and PowerPoint skills, which mostly teach Claude a method, this one is a guide to the right tool for each job. Its main file maps tasks to tools, with deeper material in two side files Claude opens only when needed: reference.md for advanced options and forms.md for form filling.

Task	Tool the skill points to
Merge or split PDFs	pypdf, or qpdf on the command line
Extract text	pdfplumber, or pdftotext
Extract tables	pdfplumber extract_tables()
Create a PDF	reportlab
OCR a scanned PDF	pytesseract, after converting pages to images
Fill a form	the step-by-step process in forms.md

Form filling is the most structured part. Claude first runs a script to check whether the PDF has real fillable fields. If it does, a second script lists every field with its page, position and type, and Claude fills them. If it does not, Claude works out where text should go, either from the page structure or by converting pages to images and estimating positions, then checks the boxes before writing and checks the result afterwards. Eight helper scripts in the folder cover these steps.

Is it for you?

Good fit if PDFs arrive in your inbox and you retype data from them, or if you send out forms, reports or certificates as PDFs. It suits admin, operations, research and finance work where the PDF is the format you cannot avoid.

It will not make a scanned document perfect. OCR quality depends on the scan, so check numbers before you rely on them.

What you can do with it

1. Table to spreadsheet. “Extract the tables on pages 3 to 5 of this annual report into a CSV.” Pair it with the xlsx skill if you want a formatted workbook.
2. Fill a form. “Fill this registration form with the details in my notes. Show me the filled form before you finish.” The skill renders pages to images so Claude can check its work.
3. Batch merge. “Combine these 12 invoices into one PDF in date order.”
4. Watermark or protect. “Add a DRAFT watermark to every page” or “Add a password to this file.” Both are listed tasks in the skill.
5. Make a scan searchable. “OCR this scanned contract and give me the text.”

How to set it up

claude.ai or the Claude desktop app

1. In Settings > Capabilities, turn on Code execution and file creation.
2. In Customize > Skills, the document skills appear under From Anthropic & Partners. They turn on automatically for matching requests.
3. Attach your PDF and say what you want done with it.

On Team and Enterprise plans, an Owner enables code execution and skills under Organization settings > Capabilities first.

Claude Code

1. Run /plugin marketplace add anthropics/skills
2. Run /plugin install document-skills@anthropic-agent-skills and choose a scope.
3. Ask: “Use the PDF skill to extract the form fields from forms/application.pdf.” That is the example prompt in the anthropics/skills README.

Claude Code’s plugin details put this skill at about 160 tokens always loaded and about 3k tokens when used. Both install commands completed in our check on Claude Code 2.1.302 (2026-10-18); we did not process a PDF in that check.

Watch out for

• License. Source-available, not open source. Anthropic’s LICENSE.txt in the folder prohibits copying it outside Anthropic’s services, making derivative works and redistributing it.
• Libraries have their own licenses. The skill’s reference file lists them: pypdf (BSD), pdfplumber (MIT), reportlab (BSD), qpdf (Apache), poppler-utils (GPL-2), pdf-lib (MIT). This matters only if you reuse those tools in your own product.
• Local installs. In Claude Code, the Python libraries and command-line tools above must be present on your machine. OCR also needs pytesseract and pdf2image.
• Forms without fields. When a PDF has no fillable fields, Claude places text by coordinates. Always look at the filled pages before you submit anything official.
• Sensitive documents. In claude.ai, files you upload are processed in Claude’s sandbox under your plan’s data terms. Remove ID numbers or bank details you do not need Claude to see.

Alternatives

• Excel spreadsheets skill once the data is out of the PDF.
• Word documents skill if you need an editable document rather than a fixed PDF.
• Google Drive connector, which can read PDFs stored in your Drive without uploading them.