Claude PDF Skill: Read, Create, and Edit PDFs

The claude pdf skill is really two capabilities. First, Claude can natively read a PDF you upload — pulling both its text and its visual layout. Second, the document PDF skill lets Claude create, fill, split, and edit PDF files. Reading is built in; creating and manipulating comes from a Skill.

PDF reading behaviour and Skill availability verified 31 July 2026 against Anthropic's PDF support documentation.

That distinction trips a lot of people up, so it is worth getting straight before you do anything else. If all you want is to summarize a report or pull numbers out of a document, you do not need to enable anything — just upload the file and ask. If you want Claude to hand you back a finished PDF, or fill in a form, that is where the claude pdf skill (the document Skill) comes in. Claude currently spans models like Opus 5 and Sonnet 5, and the split described here holds across them — we review this guide regularly, and you can confirm the reading half yourself in under 30 seconds by dropping a PDF into a chat. This guide covers both sides, plus the limits worth knowing.

What the Claude PDF skill actually means

Two separate things share the name. Native PDF support renders each page as an image and extracts its text, on by default. The document Skill is a folder of instructions and scripts Claude loads only when a request needs file manipulation.

People search for the Claude PDF skill meaning slightly different things, so let us name them clearly.

Native PDF support (reading). When you attach a PDF, Claude converts each page into an image and also extracts the page text, then analyzes both together. That means it can read a clean, text-based PDF and also make sense of a scanned or image-heavy one using vision — charts, diagrams, signatures, tables, and layout included. This is the same underlying strength behind Claude's vision features, and it is on by default in the apps and the API.

The document PDF skill (creating and manipulating). Claude on its own understands PDFs but is limited in its ability to change them — it cannot, by itself, fill a form field or stitch two files together. The PDF Skill closes that gap. A Skill is a small folder containing a SKILL.md file with instructions plus, in this case, pre-written scripts. When your request calls for it, Claude loads the skill and runs those scripts to produce or edit a real PDF. If you are new to the idea, our explainer on Claude Code skills walks through how Skills work in general.

So: reading a PDF is native. Creating or editing one is the claude pdf skill doing the heavy lifting.

What you can do with PDFs in Claude

Summarizing, question-answering, table extraction, translation, and document comparison all run on native reading. Filling forms, generating a formatted PDF, and merging or splitting files need the document skill. The dividing line is whether you want a file back.

Here is a quick map of common jobs and how Claude approaches each. Some rely on native reading; others need the document skill.

The rule of thumb: if the answer is "tell me about this PDF," native reading handles it. If the answer is "give me a PDF back" or "change this PDF," you want the skill.

How to read a PDF in Claude

Attach the file and ask in plain English — there is nothing to set up. Because Claude sees rendered pages alongside extracted text, scanned invoices and bar charts work as well as clean body text. Spot-check any number that matters.

This part is genuinely simple. In the Claude apps, attach the PDF to your message and ask your question in plain English — "summarize the key risks in this contract," "what was Q3 revenue," "pull every line item into a table." Because Claude sees both the text and the rendered page, it can answer about a bar chart or a scanned invoice just as well as clean body text. This makes it a strong companion for data analysis work where numbers live inside dense reports.

For scanned or photographed documents, the vision layer does the OCR-style reading for you — no separate tool needed. Anthropic's PDF support documentation explains that each page is turned into an image and its text is extracted alongside, which is why visual elements come through.

A few practical tips for reading:

• Ask specific questions. "Extract the payment schedule as a table" beats "tell me about this."
• For multi-document tasks, upload all the files, then ask Claude to compare or reconcile them.
• Always spot-check extracted numbers against the source. Claude is accurate, but on dense financial tables a quick verification is smart.

While pressure-testing this guide, we dropped a few of our own multi-column reports into a chat and asked Claude to pull specific figures into a table. The clean, text-based PDFs came back tidy on the first try; a low-quality scan we tried needed one more, more pointed prompt before the right column lined up. That two-pass rhythm — ask, then confirm the single number that actually feeds a decision — is now how we read any PDF here by default.

How to use the claude pdf skill to create and edit

Describe the outcome, not the operation. Claude recognizes that a request needs file manipulation, loads the skill, and runs its bundled scripts to extract, generate, merge, split, rotate, watermark, fill forms, or encrypt. You never name the script.

Creating or manipulating a PDF is where the document skill earns its keep. The idea is that Claude, when it recognizes your request needs file manipulation, reaches for the PDF skill and runs its bundled scripts — for example, a script that reads a PDF and lists every form field, so Claude can fill it without loading the whole file into context.

Where can you use it? Skills are available across Anthropic's surfaces, and availability differs by product and plan, so treat the specifics as evolving:

• In the Claude apps, where Skills are supported, the document skills that power PDF, Word, Excel, and PowerPoint output are available so Claude can hand back finished files.
• Via the API, you can enable Skills for your own agents and workflows.
• In Claude Code and developer setups, you can add the PDF skill or write your own. Anthropic publishes the source for its document skills as a reference.

The interaction is conversational. You say something like "fill this registration form with the details below and give me the PDF," or "turn this outline into a clean, formatted PDF report," and Claude uses the skill to produce the file. Under the hood the PDF skill can extract text and tables, create PDFs from scratch, merge and split documents, rotate pages, add watermarks, handle forms, and even encrypt files — but you never have to name those operations. You describe the outcome; the skill picks the right script. Its siblings work the same way, so the Claude PowerPoint skill builds an editable .pptx deck from a brief with the same conversational approach.

If you want to go further and build a PDF workflow tailored to your own templates, our walkthrough on how to build a Claude skill shows the anatomy of a SKILL.md file, and our roundup of the best Claude skills is a good place to see what others have shipped. The official Agent Skills documentation is the canonical reference for how skills load and run.

When you need the skill vs. when you do not

Skip it for summarizing, translating, answering questions, extracting text for a quick answer, or comparing two uploads. Reach for it when you want a brand-new file back, a completed fillable form, or restructured existing PDFs.

Because this is the single most common point of confusion, here is the plain version. You do not need the claude pdf skill to:

• Summarize, translate, or answer questions about a PDF.
• Extract text or read a table for a quick answer.
• Compare two uploaded documents.

You do want the claude pdf skill to:

• Get a brand-new PDF file back from Claude.
• Fill out a fillable form and receive the completed PDF.
• Merge, split, watermark, or otherwise restructure existing PDFs.

If you find yourself thinking "I want a file, not just an answer," that is your signal that the skill is involved.

Limits and good practice

Four realities worth knowing: file-size and page caps that shift over time, degraded accuracy on faint or skewed scans, extraction that deserves verification whenever a number feeds a decision, and the privacy weight of uploading documents at all.

A few realities keep this feature honest, and knowing them saves frustration.

Size and page limits exist. Both the apps and the API cap how large a file can be and how many pages Claude will fully analyze with vision; beyond a certain length, only text is processed. The exact numbers shift as models improve, so check the current official PDF support docs rather than trusting a figure you read somewhere a year ago. If a document is huge, split it and feed the relevant sections.

Scanned fidelity varies. Vision reads scanned pages well, but a faint fax or a skewed photo can lose accuracy. Clean scans in, clean data out.

Verify the numbers. For anything that feeds a decision — invoices, financial tables, legal figures — treat Claude's extraction as a fast first pass and confirm the values against the source.

Mind privacy. PDFs often carry sensitive personal or financial data. Only upload documents you are comfortable processing, and follow your organization's data policies. If you are still getting oriented on the platform generally, our primer on what Claude AI is covers the basics of how it handles your content.

Used with those guardrails, the claude pdf skill turns PDF work from a chore into a conversation — read them, question them, and get finished files back without leaving the chat.