SylphxAI/pdf-reader-mcp

Give your AI agent eyes for PDFs — structured text, tables, OCR, visual evidence, and page-level citations via MCP. Native Rust, local-first.

What it solves

PDF Reader MCP solves the problem of AI agents hallucinating or losing context when reading PDFs. Instead of dumping plain text, it provides structured data, including tables, OCR for scanned documents, and precise page-level citations, ensuring agents have verifiable evidence for their answers.

How it works

It operates as a Model Context Protocol (MCP) server that provides three primary tools to an AI agent:

  • read_pdf: Extracts markdown, tables, and structure with OCR and citations.
  • search_pdf: Locates specific page and snippet matches before performing a deep read.
  • pdf_evidence: Handles focused evidence operations, including cropping and rendering.

The system uses a native Rust engine for high performance and a thin Node.js launcher to integrate with MCP clients like Claude Desktop, Cursor, or VS Code.

Who it’s for

This tool is designed for developers building AI agents that need to process complex documents such as financial reports, research papers, and scanned PDFs where accuracy and provenance are critical.

Highlights

  • Structured Extraction: Converts PDFs into an "Agent Document Twin" featuring markdown, geometry, and provenance.
  • High Performance: Uses a native Rust engine that offers up to 10x median latency improvement over previous TypeScript versions.
  • Local-First: Processes documents locally on the supported platforms (macOS, Linux, Windows).
  • Low Footprint: Significantly reduces the installed file count and dependency graph compared to traditional JS-based PDF tools.