The Claude PDF skill: what it does and how to install it

Source: anthropics/skills · Maintainer: Anthropic · License: Proprietary (source-available) · Checked: 23 October 2026

What is the Claude PDF skill?

The Claude PDF skill is Anthropic's Agent Skill for doing real work with PDF files: pulling out text and tables, merging and splitting, filling forms, running OCR on scans and building new PDFs. It is for anyone who wants Claude to produce or change an actual .pdf file rather than describe what to do.

It is one of four document skills (with docx, xlsx and pptx) that the repo README says "power Claude's document capabilities." Its SKILL.md description reads: "Use this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill."

That description is broad on purpose. Claude reads only the name and description of each skill at the start of a session, and loads the full instructions when a request matches, as Anthropic's Agent Skills overview explains. Mentioning a .pdf file is enough to trigger it.

What does the PDF skill do?

The SKILL.md is a practical guide to Python libraries and command-line tools, with short code recipes Claude adapts to your task:

• pypdf for merging, splitting, rotating pages, reading metadata, watermarking and adding passwords.
• pdfplumber for text with layout and for table extraction, with a recipe that combines tables into a pandas DataFrame and saves them to Excel.
• reportlab for creating PDFs, from a single page to multi-page reports. It warns never to use Unicode subscript or superscript characters, because the built-in fonts render them as black boxes, and to use <sub> and <super> tags instead.
• Command-line tools: pdftotext and pdfimages from poppler-utils, qpdf for merging, splitting, rotating and decrypting, and pdftk if it is installed.
• OCR for scanned PDFs with pytesseract and pdf2image.

The folder holds more than the one file:

• forms.md: a step-by-step form-filling process. It starts by checking whether the PDF has fillable fields. If not, Claude finds field positions from the page structure (preferred) or from page images, validates the bounding boxes, then fills the form and checks the output.
• reference.md: advanced material on pypdfium2, the JavaScript libraries pdf-lib and pdfjs-dist, and more poppler and qpdf options.
• scripts/: eight Python helpers, including check_fillable_fields.py, extract_form_field_info.py, extract_form_structure.py, fill_fillable_fields.py, fill_pdf_form_with_annotations.py, convert_pdf_to_images.py, check_bounding_boxes.py and create_validation_image.py.

Because scripts run through bash and only their output enters the context, the form workflow uses fewer tokens than having Claude write the same code each time.

How do you install the PDF skill?

There are several documented routes. You only need one.

In the Claude app. The PDF skill is built in. Anthropic's Use skills in Claude article says skills are available on Free, Pro, Max, Team and Enterprise plans and need code execution. On an individual plan, turn on Code execution and file creation in Settings, Capabilities, then manage skills in Customize, Skills. Claude uses the PDF skill automatically when relevant.

In Claude Code, signed in with your claude.ai account. The Claude Code skills docs say terminal sessions signed in with a claude.ai account sync the account's skills into ~/.claude/skills/synced/ (Claude Code v2.1.273 or later), and that "some of Anthropic's skills, such as pdf and xlsx, always sync." Run /skills and look under claude.ai sync. This doesn't happen when you sign in with an API key or use Amazon Bedrock. Anthropic's Agent Skills overview still says the pre-built document skills are not available in Claude Code, so if pdf doesn't appear, use the plugin route below.

As a Claude Code plugin. The repo README registers the repo as a plugin marketplace. In a Claude Code session:

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills

The document-skills plugin installs all four document skills at once. Plugin skills are namespaced, so you can also call it directly as /document-skills:pdf.

With the skills CLI. The skills.sh page gives this command:

npx skills add https://github.com/anthropics/skills --skill pdf

Add -a claude-code to target Claude Code. Without -g the folder lands in your project at .claude/skills/pdf/; with -g it goes to ~/.claude/skills/pdf/. Codex and Cursor read the same SKILL.md format (see Codex skills and Cursor skills), but read the license section below before using these files outside Anthropic's products.

Through the API. The pre-built skill has the skill_id pdf and runs with the code execution tool; see the overview.

How do you use the PDF skill?

Just describe the job and name the file. The README's own example is "Use the PDF skill to extract the form fields from path/to/some-file.pdf". Some other requests that match the skill:

• "Merge invoice-jan.pdf, invoice-feb.pdf and invoice-mar.pdf into one file, in that order."
• "Pull every table out of annual-report.pdf and save them to a spreadsheet."
• "This scanned contract has no text layer. OCR it and give me a searchable copy."
• "Fill in this W-9 with the details below and save it as w9-filled.pdf."
• "Add a CONFIDENTIAL watermark to every page and password-protect the result."

Requirements on your own machine. The skill expects Python with pypdf, pdfplumber and reportlab (plus pandas for the table recipe), and for OCR pytesseract and pdf2image. The command-line recipes need poppler-utils and qpdf; pdftk is optional. In the Claude app these run in Claude's sandbox, so you install nothing.

Tips and limits

• It is not open source. The README says the document skills "are source-available, not open source." The folder's LICENSE.txt ties use to your agreement with Anthropic (the Consumer or Commercial Terms) and adds restrictions: no extracting the materials from the Services or keeping copies outside them, no reproducing, no derivative works, no distribution. You can read it and learn from it, but don't fork it into your own edited version. If you want different behavior, write your own skill.
• The repo copy may not match what Claude runs. The README warns that "the implementations and behaviors you receive from Claude may differ from what is shown in these skills."
• It runs code on your files. In Claude Code, scripts execute on your computer with your permissions. Anthropic's overview advises using skills only from trusted sources; Anthropic is the maintainer here, but still review what Claude proposes to run on sensitive documents.
• Form filling is careful work. forms.md tells Claude to finish the steps in order and validate bounding boxes before writing anything, so expect a few passes on a complex form.
• Check the output. Open the final PDF yourself, especially after OCR or form filling.

Save it in your Kipt library

Because the license forbids copying these files, don't import the pdf folder itself into a library. What you can keep in Kipt is your own work around it: the PDF prompts you reuse, with {{variables}} for file names and fields, and your own skills, which you can draft with the free SKILL.md generator. Kipt stores them as Markdown in your own private GitHub repo, and one command installs every skill into Claude Code or Codex:

npx skills add https://github.com/<you>/<repo>.git --all -g

Similar skills

• Claude docx skill for Word documents.
• Claude Excel skill for spreadsheets and CSV files.
• Claude PowerPoint skill for slide decks.
• Best Claude Code skills for the wider list, including Anthropic's open-source example skills.