PDF (By Anthropic) MCP Server

Anthropic's PDF Agent Skill for reading, extracting, creating, merging, splitting, watermarking, OCR, and form-filling PDF documents.

Developer Tools by Anthropic None active

GitHub Docs

Overview

PDF is an official Anthropic Agent Skill (distributed via the anthropics/skills repository) that teaches Claude how to perform a wide range of PDF operations. It is not a traditional MCP server with a fixed tool list; it is a skill package (a SKILL.md file plus reference docs and helper scripts) that Claude loads dynamically to handle PDF related tasks. In Claude Code it is installed as part of the document-skills plugin, which sits in the same plugin/MCP ecosystem alongside other connectors.

The skill covers essentially the full PDF lifecycle: extracting text and tables (via pdfplumber), basic operations like merging, splitting, rotating, and metadata edits (via pypdf), authoring new PDFs (via reportlab), OCR on scanned documents (via pytesseract and pdf2image), image extraction (pdfimages), watermarking, password protection, and PDF form filling (covered in a dedicated FORMS.md reference). It also documents command line equivalents using qpdf, pdftk, and pdftotext.

What makes it notable is that it is the same skill family that powers Claude's production document features on Claude.ai paid plans, source-available rather than fully open source. Because it works by instructing Claude how to use existing Python and CLI tools rather than running a hosted service, it works fully offline once dependencies are installed.

Tools

Tool	Description
`extract_text`	Extract plain or layout-preserving text from PDFs using pdfplumber or pdftotext.
`extract_tables`	Pull tables from PDFs into structured data (pandas/Excel) using pdfplumber.
`merge_pdfs`	Combine multiple PDFs into one document using pypdf or qpdf.
`split_pdf`	Split a PDF into individual pages or page ranges.
`rotate_pages`	Rotate one or more pages by 90/180/270 degrees.
`add_watermark`	Overlay a watermark PDF onto another PDF using pypdf merge_page.
`create_pdf`	Generate new PDFs from scratch with reportlab, including multi-page Platypus layouts.
`fill_form`	Programmatically fill PDF AcroForm fields (see FORMS.md for guidance).
`encrypt_pdf`	Apply password protection / encryption to a PDF.
`decrypt_pdf`	Remove password protection from a PDF using qpdf.
`extract_images`	Extract embedded images from a PDF using pdfimages.
`ocr_pdf`	Run OCR on scanned PDFs using pytesseract and pdf2image to produce searchable text.

Setup Guide

Prerequisites

Claude Code (or Claude.ai with paid plan, or Claude API with Skills support)
Python 3 with the libraries used by the skill: pypdf, pdfplumber, reportlab, pdf2image, pytesseract
Optional CLI tools: qpdf, pdftk, poppler-utils (pdftotext, pdfimages), tesseract

Install in Claude Code (recommended)

The PDF skill ships as part of the document-skills plugin from Anthropic's official skills marketplace:

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

Once installed, invoke it by mentioning PDF work in your prompt, for example:

Use the PDF skill to extract the form fields from ./contract.pdf

Manual install / inspection

Clone the repo to read or vendor the skill directly:

git clone https://github.com/anthropics/skills.git
cd skills/skills/pdf

Files of interest:

SKILL.md (main instructions and YAML frontmatter)
FORMS.md (PDF form filling guidance)
reference.md (extended API reference)
scripts/ (helper scripts)

Claude.ai

Built-in for paid plans. Custom variants can be uploaded under Settings > Capabilities > Skills.

Claude API

Use the Skills API to register the skill. See the Anthropic Skills guide at docs.claude.com/en/api/skills-guide.

Note

This is an Agent Skill, not a traditional standalone MCP server. It augments Claude with a SKILL.md plus helper scripts rather than exposing a long-lived JSON-RPC server.

Use Cases

Extract structured tables from invoices, bank statements, or financial reports into pandas or Excel for downstream analysis.
Auto-fill recurring PDF forms (W-9, NDAs, intake paperwork) given a JSON payload of field values.
Merge a generated cover letter, signed contract, and supporting exhibits into a single delivery-ready PDF.
Run OCR on scanned legal or medical PDFs to make them searchable, then extract specific clauses or data points.
Generate branded multi-page PDF reports with reportlab from query results or Markdown content.

Example Prompts

"Use the PDF skill to extract all tables from q3-financials.pdf and save them as separate sheets in an Excel file."
"Merge cover.pdf, contract.pdf, and exhibits.pdf into one file called signed-package.pdf."
"Fill the form fields in w9-template.pdf with the values in vendor.json and save as acme-w9.pdf."
"OCR the scanned PDF lease.pdf, then find every clause that mentions a renewal date."
"Create a 5-page PDF report from this Markdown summary, with a title page and page numbers."

Pros

Maintained by Anthropic and source-available in the official anthropics/skills repo, so it tracks what Claude actually uses in production.
Broad coverage: extraction, creation, merging, splitting, OCR, encryption, watermarks, and form filling in one skill.
Works fully locally using standard Python and CLI tooling, with no external API calls or cloud costs.
Includes dedicated supplementary docs (FORMS.md, reference.md) for the tricky cases.

Limitations

It is an Agent Skill, not a long-running MCP server, so it does not expose a fixed JSON-RPC tool surface and behavior depends on Claude following the instructions.
License is proprietary ("source-available"), so it cannot be freely redistributed or modified like an MIT/Apache MCP server.
Requires Python plus optional CLI dependencies (tesseract, qpdf, poppler) to be installed on the host for full functionality.

Alternatives

Markdownify MCP: converts PDFs, DOCX, PPTX, HTML, and YouTube transcripts into Markdown via an MCP server.
Docling MCP: IBM's document parsing toolkit with a community MCP server for PDF to Markdown conversion.
Desktop Commander: general-purpose MCP server with file operations across PDF, DOCX, Excel, and more.