AGENT SKILL

Stop wasting your context window on PDF extraction

Extracting Markdown from PDFs takes precision. Nutrient's pdf-to-markdown skill gives Claude and Codex fast, accurate extraction, so your AI workflow starts with clean Markdown instead of noisy output that eats into your context window.

Fast and accurate

91x faster than docling, with reading order and extraction accuracy that lead the field.

Your PDFs stay local

Runs on your machine — no documents are uploaded to Nutrient.

Works with Claude Code and Codex

Install via the marketplace, use it in Codex, or run the standalone CLI — free for up to 1,000 documents a month.

INSTALL

Install via the plugin marketplace, or use the CLI

Two ways to get the skill on the machine: through an agent's plugin marketplace, or as a standalone binary.

/

Install it in Claude Code or Codex

We published pdf-to-markdown in the Nutrient marketplace. Claude Code and Codex both read from the same marketplace, so the commands below install it in either one.

/plugin marketplace add pspdfkit-labs/nutrient-skills
/plugin install pdf-to-markdown@nutrient-skills

Install the standalone CLI

Or install the standalone pdf-to-markdown CLI tool directly. Requires macOS or Linux; verify with pdf-to-markdown --version. Inspect the install script on GitHub.

curl https://agent-cdn.nutrient.io/pdf-to-markdown/install.sh | sh

THE NUTRIENT DOCUMENT CLI

One local binary, three commands

The same signed binary that powers pdf-to-markdown also extracts layout-preserving plain text and searches what you already extracted. All local, all on the same free tier.


pdf-to-markdown

Structured Markdown

Clean Markdown with headings, lists, tables, and reading order. The right input for most RAG and LLM-context pipelines.


pdf-to-text

Layout-preserving text

Plain text that keeps columns and tabular alignment, for plain-text models and column-sensitive pipelines.


query

Search what you extracted

Ranked search over an extracted file, returning only the top line windows. Parse once, then ask as many questions as you like.


BENCHMARKS

Speed and accuracy shouldn't be a tradeoff

Better PDF-to-Markdown extraction means less cleanup, fewer wasted tokens, and more reliable downstream results for Claude and Codex.

Nutrient Other tools
Where Nutrient lands on quality
Reading order (NID)
0.92 #1 OF 7
Extraction accuracy
0.88 TOP 2
Heading level (MHS)
0.81 TOP 2
ACCURACY ↑ Higher is better · Score 0–1
Extraction accuracy
docling 0.88
Nutrient 0.88
opendataloader 0.83
pymupdf4llm 0.82
markitdown 0.59
pypdf 0.58
liteparse 0.57
Reading order (NID)
Nutrient 0.92
opendataloader 0.90
docling 0.90
pymupdf4llm 0.88
pypdf 0.87
liteparse 0.86
markitdown 0.84
Table structure (TEDS)
docling 0.89
Nutrient 0.66
opendataloader 0.49
pymupdf4llm 0.48
markitdown 0.27
liteparse 0.00
pypdf 0.00
Heading level (MHS)
docling 0.82
Nutrient 0.81
pymupdf4llm 0.77
opendataloader 0.74
markitdown 0.00
liteparse 0.00
pypdf 0.00
SPEED ↓ Lower is better · And it isn't close
Extraction time per page Seconds · Log scale
Nutrient 0.007s
opendataloader 0.014s
pypdf 0.019s
markitdown 0.106s
liteparse 0.233s
pymupdf4llm 0.252s
docling 0.618s

91×
faster than docling
37×
faster than pymupdf4llm
34×
faster than liteparse
16×
faster than markitdown
faster than pypdf
faster than opendataloader

Benchmark data last updated April 2026, recorded on Apple M3 Ultra (no GPU). See the GitHub repository for methodology and the latest runs.

These are born-digital PDFs, benchmarked against open source parsers. Nutrient leads on reading order and speed; docling leads on complex table structure (0.89 vs. 0.66 TEDS). For scanned or handwritten documents, and for schema-level extraction with per-value confidence, use the Data Extraction API. Full methodology is in the benchmark documentation.

SECURITY PDFs don't go to Nutrient

The skill reads PDFs and produces Markdown output without uploading documents to Nutrient. Claude, Codex, and any other model provider still apply their own policies.

Document privacy illustration with a shield and eye icon

LICENSING

Free to use, with clear limits

Use the extractor freely within the included usage limits, then move to a commercial license as you scale.

FREE FOREVER

Personal

Up to

1,000

documents / month


Free for processing up to 1,000 documents per calendar month.

Start for free

Commercial

Over

1,000+

documents / month


Commercial license required for: processing more than 1,000 documents per month, redistributing the binary, or OEM/white label use.

Contact sales

Frequently asked questions

What is the PDF-to-Markdown skill for Claude and Codex?

It’s an agent skill that extracts clean, accurate Markdown from PDFs for Claude and Codex. Instead of feeding noisy PDF output into your AI workflow and burning context-window tokens on cleanup, the skill returns structured Markdown your model can work with right away.

How do I install the PDF-to-Markdown skill in Claude Code?

We published pdf-to-markdown in the Nutrient marketplace, so you can add it directly in Claude Code. Run “/plugin marketplace add pspdfkit-labs/nutrient-skills”, and then run “/plugin install pdf-to-markdown@nutrient-skills”.

Does it work with Claude Code and Codex?

Yes. You can install it in Claude Code through the Nutrient marketplace, use it in Codex, or install the standalone CLI directly with “curl https://agent-cdn.nutrient.io/pdf-to-markdown/install.sh | sh”.

How is this different from passing a PDF directly to Claude?

Passing a raw PDF to a model produces noisy extraction that wastes tokens and degrades downstream results. The skill returns fast, accurate Markdown with strong reading order and table structure, so your AI workflow starts from clean input instead of cleaning up messy output.

What does it cost, and do my PDFs go to Nutrient?

It’s free for processing up to 1,000 documents per calendar month. A commercial license is required for processing more than 1,000 documents per month, redistributing the binary, or OEM/white-label use. Your PDFs aren’t uploaded to Nutrient — the skill reads them and produces Markdown without routing your documents through a Nutrient service.

Can I use it with other LLM agents?

Yes. The skill runs in Claude Code, in Codex, or as a standalone CLI, so it fits into other agent workflows too. Whichever model provider you use still applies its own policies, and your local PDFs stay inside your workflow.

What’s the difference between pdf-to-markdown, pdf-to-text, and query?

They’re commands backed by the same local binary. pdf-to-markdown returns structured Markdown (headings, lists, tables), which is what most RAG and LLM-context pipelines want. pdf-to-text returns layout-preserving plain text for plain-text consumers and column-sensitive pipelines. query runs ranked search over a file you already extracted, so an agent can pull one fact without reading the whole document back into context.

What about scanned or handwritten PDFs?

This CLI is built for born-digital PDFs that already contain a text layer, and it doesn’t OCR. For scanned, photographed, or handwritten documents, and for schema-level structured extraction with per-value coordinates and confidence, use the Nutrient Data Extraction API, which is designed for exactly that.

MORE FROM NUTRIENT AI

Part of the Nutrient document AI platform

The PDF-to-Markdown skill is one piece — more agent skills are on the way. Explore the rest of the platform.