pdf-mcp

Agentic RAG over one PDF or a whole folder, with page-level evidence to check every answer: hybrid search, selective page reads, tables, OCR, charts, and CJK la

What is pdf-mcp?

Agentic RAG over one PDF or a whole folder, with page-level evidence to check every answer: hybrid search, selective page reads, tables, OCR, charts, and CJK layouts.

Overview of pdf-mcp

What it is

pdf-mcp is an MCP (Model Context Protocol) server that allows AI agents like Claude Code to read and search PDFs. It supports single files or entire folders, performing agentic RAG with hybrid semantic and keyword search. It can extract tables, images, OCR text, chart data, and handle multi-column and CJK (Chinese, Japanese, Korean) layouts. It uses a SQLite cache to avoid re-parsing and optionally leverages CUDA for faster embedding on NVIDIA GPUs.

Key features

  • Hybrid search combining BM25 keyword and semantic embedding for relevant excerpts.
  • Selective page reading: read only specific pages or ranges, with OCR on demand.
  • Table and chart data extraction: returns structured rows and (x, y) data from vector charts.
  • Multi-column and vertical layout support, including Japanese tategaki (vertical writing).
  • OCR for scanned PDFs via Tesseract, parallel across pages.
  • Folder-level tools: warm a corpus, get per-document overview, search across all PDFs.
  • SQLite cache that persists across restarts, with cache statistics and clearing tools.

Use cases

  • Summarizing large documents (e.g., 200-page annual reports) by first getting page info, searching for relevant sections, then reading only needed pages.
  • Searching across a folder of PDFs for specific topics or keywords.
  • Extracting tables and chart data from reports for analysis.
  • Reading scanned or multi-column documents with correct reading order.
  • Integrating with AI agents to answer questions about PDF content without loading entire documents.

Getting started

Requires Python 3.11 or later. Install the pdf-mcp command using uv or pipx: bash uv tool install pdf-mcp # or: pipx install pdf-mcp For Claude Code, add the MCP server: bash claude mcp add pdf-mcp -- pdf-mcp For Claude Desktop, download the .mcpb file and drag it into Settings > Extensions. See the repository README for other clients.

License and pricing

MIT License, open source. No pricing information is stated beyond being open source.

Based on the project's public README and repository metadata.

Add this badge to your README

Featured on YunoDB Badge styles & help

Maintain this project? Show that it is featured on YunoDB.

Markdown[![Featured on YunoDB](https://yunodb.com/badge/jztanpdf-mcp.svg)](https://yunodb.com/tools/jztanpdf-mcp?utm_source=github&utm_medium=badge)
HTML<a href="https://yunodb.com/tools/jztanpdf-mcp?utm_source=github&utm_medium=badge"><img src="https://yunodb.com/badge/jztanpdf-mcp.svg" alt="Featured on YunoDB" height="20"></a>