SylphxAI/pdf-reader-mcp

What it solves

Citra prevents AI agents from hallucinating when reading PDFs. While most PDF tools simply dump plain text—causing agents to lose table structures, miss page numbers, or miscite data—Citra provides an "Agent Document Twin" that includes structured text, tables, OCR, and precise geometry for verifiable citations.

How it works

It uses a native Rust engine (delivered via a thin Node launcher) to parse PDFs locally. It converts documents into markdown with preserved structure and provenance, allowing agents to reference specific pages, tables, and cells. It is available as an MCP (Model Context Protocol) server, a CLI, or a TypeScript SDK.

Who it’s for

Developers building AI agents that need to process complex documents like financial reports, research papers, or scanned PDFs where accuracy and citeable evidence are critical.

Highlights

• Evidence-Based Output: Returns precise coordinates, page numbers, and table cell references instead of just text.
• Local-First: Processing happens locally on the user's machine across five major platforms (macOS, Linux, Windows).
• High Performance: Features a Rust-based engine that offers significantly lower latency and a smaller installation footprint compared to JavaScript-based alternatives.
• Versatile Integration: Supports MCP for easy connection to clients like Claude Desktop and Cursor, as well as a programmatic SDK.